Distributed energy consumption optimization method and system based on multi-modal data synchronization and generation

Abstract

The application provides a kind of based on multimodal data synchronization and generation distributed energy consumption optimization method and system, it is related to energy management technical field, the method includes: the multivariate energy consumption time series data of controlled energy-using equipment is collected, and multivariate energy consumption time series data is processed to abnormal value;The multivariate energy consumption time series data after processing is stored in distributed manner, and the multivariate energy consumption time series data stored in distributed manner is dynamically synchronized with the physical response lag feature of controlled energy-using equipment;Multivariate energy consumption time series data after dynamic synchronization is processed is determined by clustering algorithm, Newton iteration method to abnormal working condition data;Multivariate energy consumption time series data is processed using pre-constructed adversarial network pair, to obtain authenticity discrimination probability and future energy consumption prediction value;Determine target control parameter based on future energy consumption prediction value;Combined with physical response lag feature, authenticity discrimination probability, target control parameter is sent to controlled energy-using equipment for regulation and control asynchronously.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of energy management technology, and in particular to a distributed energy consumption optimization method and system based on multimodal data synchronization and generation. Background Technology

[0002] Distributed integrated energy management networks typically integrate controlled energy-consuming terminals covering various physical media such as electricity, natural gas, and water resources. To improve overall energy efficiency and reduce losses, the control center needs to acquire operating parameters of underlying devices through IoT sensor arrays deployed on-site. Then, it uses various predictive algorithms to output future load trends and issue scheduling commands for overall regulation. Under complex industrial operating conditions, time-series data collected by multi-source sensor channels are often affected by electromagnetic harmonics or changes in contact impedance, resulting in discrete and distorted values. Monitoring channels for different media also objectively exhibit clock drift due to independent hardware operation. Directly using such data sequences without strict time alignment and anomaly smoothing for energy consumption prediction can easily lead to misalignment of the final optimization decisions on the time axis.

[0003] Controlled hardware units with different physical properties exhibit objective physical hysteresis differences in their state response phase after receiving control parameters. Purely electrical devices such as photovoltaic inverters and energy storage charge / discharge controllers can quickly achieve power-level response and follow after receiving level signals. Fluid devices such as hydraulic control units and gas mechanical valves need to overcome inherent static friction dead zones and medium flow inertia when their actions change, and their physical state transition often requires a relatively long ramp-up period. Current energy collaborative scheduling systems mostly adopt synchronous triggering logic when issuing multi-dimensional optimization parameters on the same time reference plane. If the response rate differences between underlying device groups are not fully considered, once the system faces a significant demand for operating condition switching, fast-responding electrical components will instantly switch to the expected state, while slow-responding mechanical fluid components will still be lagging behind. The timing disconnect at the underlying physical execution level can easily cause short-term local energy accumulation and state mismatch in the pipeline network or electrical circuit, making it difficult to ensure the coordinated matching of operating conditions of multi-media devices during the transition phase. Summary of the Invention

[0004] To address the aforementioned technical problems, embodiments of the present invention provide a distributed energy consumption optimization method based on multimodal data synchronization and generation, comprising: Collect multidimensional energy consumption time-series data of controlled energy-consuming devices in a distributed integrated energy management and control network, and perform outlier processing on the multidimensional energy consumption time-series data; The processed multidimensional energy consumption time series data is distributed and stored in a distributed manner. The distributed stored multidimensional energy consumption time series data is then dynamically synchronized in combination with the physical response lag characteristics of the controlled energy-consuming equipment. Clustering algorithms and Newton's iteration method are used to determine abnormal operating condition data for the multidimensional energy consumption time series data after dynamic synchronization processing; By using a pre-built adversarial network to process the multidimensional energy consumption time series data that has been dynamically synchronized and includes the abnormal operating condition data, the authenticity judgment probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment are obtained. Based on the predicted future energy consumption values, target control parameters for regulating the controlled energy-consuming equipment are determined. Combining the physical response hysteresis characteristics and the authenticity discrimination probability, the target control parameters are asynchronously sent to different types of controlled energy-consuming devices to regulate the controlled energy-consuming devices so that the action variation curve of the fast-response type controlled energy-consuming device is fitted with the hysteresis action curve of the slow-response type controlled energy-consuming device.

[0005] In one embodiment, the outlier processing of the multidimensional energy consumption time-series data includes: The sliding window method and the interquartile range method are used to detect outliers in the multidimensional energy consumption time series data corresponding to different sliding windows. The interquartile range multiplier is set differently for different types of acquisition channels. Calculate local variance based on sampled data before and after outliers; Based on the relationship between local variance and a preset steady-state threshold, it is determined whether the controlled energy-consuming equipment is in a stable or abrupt stage. The outliers in the stationary phase are replaced by the mean, and the outliers in the abrupt change phase are replaced by the median.

[0006] In one embodiment, the dynamic synchronization processing of the multi-dimensional energy consumption time-series data in distributed storage combined with the physical response lag characteristics of the controlled energy-consuming equipment includes: Extract the timestamp parameter from the multidimensional energy consumption time series data in the distributed storage, and combine the timestamp parameter with the calendar to convert the timestamp parameter into a high-dimensional one-hot encoded sparse matrix to obtain the time period features; The delay step is determined based on the total length of the transmission pipeline of the underlying controlled medium of the controlled energy-consuming equipment and the medium flow velocity. The physical response hysteresis characteristics of the controlled energy-consuming equipment are constructed by combining the delay step size and the corresponding energy consumption value of the controlled energy-consuming equipment at the same historical time point. The multidimensional energy consumption time series data is reduced by combining the device ID, time period characteristics, and physical response lag characteristics of the controlled energy-consuming equipment.

[0007] In one embodiment, the step of determining abnormal operating condition data by using a clustering algorithm and Newton's iteration method on the multidimensional energy consumption time-series data after dynamic synchronization processing includes: Clustering is performed on the multidimensional energy consumption time series data after dynamic synchronization processing, and candidate abnormal operating condition data are determined by combining the clustering results and the minority class proportion threshold. The minority class proportion threshold is calculated based on the prior statistical probability benchmark of the average failure interval time of the controlled energy-consuming equipment and the expected hardware failure rate. The candidate abnormal operating condition data is verified using the Newton-Raphson iteration method. If it is determined that the candidate abnormal operating condition data violates the physical manufacturing limit of the controlled energy-consuming equipment, then the candidate abnormal operating condition data is determined to be the abnormal operating condition data.

[0008] In one embodiment, constructing the adversarial network includes: A generator network is constructed, which uses a fully connected network architecture to generate simulated sequences of multidimensional energy consumption time-series data. A discriminant prediction network is constructed using a hybrid architecture of convolutional layers and bidirectional long short-term memory layers. The discriminant prediction network is trained based on the multidimensional energy consumption time-series data and simulated sequences. The process of using a pre-constructed adversarial network to process multidimensional energy consumption time-series data that has undergone dynamic synchronization processing and includes the abnormal operating condition data, to obtain the authenticity judgment probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment, includes: The discriminant prediction network in the pre-constructed adversarial network is used to process the multidimensional energy consumption time series data that has been dynamically synchronized and includes the abnormal operating condition data, so as to obtain the authenticity discrimination probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment.

[0009] In one embodiment, the multidimensional energy consumption time-series data includes natural gas consumption, water resource consumption, and electricity consumption dimensions; The determination of target control parameters for regulating the controlled energy-consuming equipment based on the predicted future energy consumption includes: Construct a comprehensive objective function that covers the aforementioned dimensions of natural gas consumption, water resource consumption, and electricity consumption; The target control parameters are solved by using a convex optimization solver to minimize the value of the integrated objective function.

[0010] In one embodiment, combining the physical response hysteresis characteristics and the authenticity discrimination probability, the target control parameters are asynchronously sent to different types of controlled energy-consuming devices, including: Based on the physical response hysteresis characteristics, the controlled energy-consuming equipment is divided into slow-response type controlled energy-consuming equipment and fast-response type controlled energy-consuming equipment; The target control parameters are preferentially sent to slow-response controlled energy-consuming devices to achieve device regulation; The actual physical flow of the pipeline network where the controlled energy-consuming equipment is located is monitored, and when the actual physical flow exceeds a preset threshold, the target control parameters are sent to the controlled energy-consuming equipment of the fast response type, and the controlled energy-consuming equipment of the fast response type is regulated in combination with the actual discrimination probability.

[0011] In one embodiment, the target control parameters include proportional-integral-derivative control parameters; The controlled energy-consuming devices of the fast-response type are regulated based on the actual discrimination probability, including: The dynamic smoothing weights of the controlled energy-consuming devices of the fast response type are determined based on the authenticity discrimination probability. The dynamic smoothing weight is used as an attenuation multiplier to constrain the control slope of the fast-response type controlled energy-consuming equipment, and the control slope is determined based on the integral time constant in the proportional-integral-derivative control parameters.

[0012] In one embodiment, regulating the controlled energy-consuming device includes: For different controlled energy-consuming devices, the target control parameters are mapped to different actual execution variables, and the actual execution variables are regulated. Specifically, for HVAC equipment among the controlled energy-consuming devices of the aforementioned fast-response type, the actual execution variable is the compressor operating frequency or the fan speed; For the energy storage battery charge / discharge controller in the controlled energy-consuming device of the aforementioned fast-response type, the actual execution variable is the charge / discharge cutoff voltage or the charge / discharge rate limit; For photovoltaic inverters in controlled energy-consuming devices of the aforementioned fast-response type, the actual execution variable is the maximum power point tracking current reference value; For the intelligent water supply pump and gas-fired wall-mounted boiler in the controlled energy-consuming equipment of the slow response type, the actual execution variables are the opening degree of the water valve and the opening degree of the gas nozzle.

[0013] Another embodiment of the present invention also provides a distributed energy consumption optimization system based on multimodal data synchronization and generation, comprising: The data acquisition module is used to acquire multi-dimensional energy consumption time-series data of controlled energy-consuming devices in the distributed integrated energy management and control network, and to process outliers in the multi-dimensional energy consumption time-series data. The first processing module is used to perform distributed storage of the processed multidimensional energy consumption time series data, and to perform dynamic synchronization processing of the distributed storage of the multidimensional energy consumption time series data in combination with the physical response lag characteristics of the controlled energy-consuming equipment. The first determining module is used to determine abnormal operating condition data by using clustering algorithm and Newton's iteration method on the multidimensional energy consumption time series data after dynamic synchronization processing. The second processing module is used to process the multidimensional energy consumption time series data containing the abnormal operating condition data after dynamic synchronization processing using a pre-built adversarial network, and to obtain the authenticity judgment probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment. The second determining module is used to determine the target control parameters for regulating the controlled energy-consuming equipment based on the future energy consumption prediction value. The control module is used to combine the physical response hysteresis characteristics and the authenticity discrimination probability to asynchronously send the target control parameters to different types of controlled energy-consuming devices, so as to control the controlled energy-consuming devices and fit the action variation curve of the fast-response type controlled energy-consuming device with the hysteresis action curve of the slow-response type controlled energy-consuming device.

[0014] Based on the above, it can be seen that the distributed energy consumption optimization method and system based on dynamic data synchronization and generative adversarial networks provided in this embodiment of the invention can effectively solve the technical problems of distributed energy collaborative control system when scheduling heterogeneous hardware devices, such as the difference in physical response hysteresis of different media causing the devices to easily experience execution disconnection and local state mismatch during the control transition phase, and the clock drift and disturbance distortion of the underlying monitoring and collection data affecting the overall optimization prediction effect.

[0015] Other features and advantages of this application will be set forth in the following description. The objectives and other advantages of this application can be realized and obtained through the structures particularly pointed out in the written description and drawings.

[0016] The technical solution of this application will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the distributed energy consumption optimization method based on multimodal data synchronization and generation in an embodiment of the present invention.

[0019] Figure 2 This is a flowchart illustrating a distributed energy consumption optimization method based on multimodal data synchronization and generation, according to another embodiment of the present invention.

[0020] Figure 3 This is a structural block diagram of a distributed energy consumption optimization system based on multimodal data synchronization and generation in an embodiment of the present invention. Detailed Implementation

[0021] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but these are not intended to limit the scope of the invention.

[0022] It should be understood that various modifications can be made to the embodiments disclosed herein. Therefore, the following description should not be considered as limiting, but merely as an example of embodiments. Other modifications within the scope of this disclosure will be apparent to those skilled in the art.

[0023] The accompanying drawings, which are included in and form part of this specification, illustrate embodiments of the present disclosure and, together with the general description of the disclosure given above and the detailed description of the embodiments given below, serve to explain the principles of the disclosure.

[0024] These and other features of the invention will become apparent from the following description of preferred forms of embodiments given as non-limiting examples, with reference to the accompanying drawings.

[0025] It should also be understood that although the invention has been described with reference to some specific examples, those skilled in the art can certainly implement many other equivalent forms of the invention.

[0026] The above and other aspects, features and advantages of this disclosure will become more apparent when taken in conjunction with the accompanying drawings and in view of the following detailed description.

[0027] Specific embodiments of the present disclosure are described thereafter with reference to the accompanying drawings; however, it should be understood that the disclosed embodiments are merely examples of the present disclosure and can be implemented in various ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that could obscure the present disclosure. Therefore, the specific structural and functional details disclosed herein are not intended to be limiting, but are merely representative of the present disclosure and are intended to teach those skilled in the art to use the present disclosure in a variety of substantially any suitable detailed structures.

[0028] This specification may use the phrases “in one embodiment,” “in another embodiment,” “in yet another embodiment,” or “in still another embodiment,” all of which may refer to one or more of the same or different embodiments according to this disclosure.

[0029] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0030] like Figure 1 As shown, this embodiment of the invention provides a distributed energy consumption optimization method based on multimodal data synchronization and generation, including: S1: Collect multi-dimensional energy consumption time-series data of controlled energy-consuming devices in the distributed integrated energy management and control network, and perform outlier processing on the multi-dimensional energy consumption time-series data; S2: Distribute the processed multidimensional energy consumption time series data and dynamically synchronize the distributed multidimensional energy consumption time series data with the physical response lag characteristics of the controlled energy-consuming equipment. S3: The abnormal operating condition data are determined by clustering algorithm and Newton's iteration method on the multidimensional energy consumption time series data after dynamic synchronization processing; S4: Using a pre-built adversarial network, the multi-dimensional energy consumption time series data, which includes the abnormal operating condition data after dynamic synchronization processing, is processed to obtain the authenticity judgment probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment. S5: Determine the target control parameters for regulating the controlled energy-consuming equipment based on the predicted future energy consumption value; S6: Combining the physical response hysteresis characteristics and the authenticity discrimination probability, the target control parameters are asynchronously sent to different types of controlled energy-consuming devices to regulate the controlled energy-consuming devices so that the action variation curve of the fast-response type controlled energy-consuming device is fitted with the hysteresis action curve of the slow-response type controlled energy-consuming device.

[0031] The distributed energy consumption optimization method and system based on dynamic data synchronization and generative adversarial networks provided in this embodiment are used to solve the technical problems of distributed energy collaborative control systems when coordinating and scheduling heterogeneous hardware devices. Due to the differences in physical response hysteresis of different media, the devices are prone to execution disconnection and local state mismatch during the control transition phase. In addition, the underlying monitoring and collection data has clock drift and disturbance distortion, which affects the overall optimization and prediction effect.

[0032] In one embodiment, the processing system for executing the distributed energy consumption optimization method based on dynamic data synchronization and generative adversarial networks of this embodiment may include, but is not limited to, a data acquisition and preprocessing module, a distributed storage and computing platform, a generative adversarial network energy consumption prediction module, a robust optimization and control module, and an abnormal behavior identification and verification unit. The modules interact with each other via a low-level industrial communication bus using feature tensors and control signaling.

[0033] The controlled energy-consuming devices are distributed within the physical space of the distributed energy system, including photovoltaic inverters, energy storage battery charge / discharge controllers, HVAC systems, gas-fired wall-hung boilers, and smart water pumps, involving multiple media such as electricity, natural gas, and water resources. The data acquisition and preprocessing module is equipped with a distributed IoT sensor array. The IoT sensor array includes voltage and current transformers for power monitoring, temperature sensors and ultrasonic flow meters for fluid monitoring, and irradiance meters for environmental parameter monitoring. The data acquisition and preprocessing module performs polling sampling on the IoT sensor array according to a set sampling frequency step size to acquire the underlying analog physical electrical signals. An analog-to-digital converter (ADC) circuit converts the analog physical electrical signals into a digital discrete signal tensor. The system processor appends a physical clock timestamp synchronized by a network time protocol and a device-specific media access control address identifier to the discrete signal tensor, and then concatenates and encapsulates it along the column dimension to form the multidimensional energy consumption time-series data matrix. For HVAC and smart water pumps in distributed microgrids, the data acquisition and preprocessing module synchronously collects the operating power of the air conditioning compressor and the chilled water circulation flow rate of the water pump circuit, and adds timestamps for the same time point. This synchronization logic constructs a multi-channel physical monitoring matrix stream aligned in the time dimension, eliminating data misalignment caused by clock drift of different media devices.

[0034] In one embodiment, the outlier processing of the multidimensional energy consumption time-series data includes: S101: The sliding window method and the interquartile range method are used to detect outliers in the multidimensional energy consumption time series data corresponding to different sliding windows. The interquartile range multiplier is set differently for different types of acquisition channels. S102: Calculate the local variance based on the sampled data before and after the outlier; S103: Based on the relationship between the local variance and the preset steady-state threshold, determine whether the controlled energy-consuming equipment is in a stable or abrupt change phase. S104: Perform mean replacement operation on outliers in the stationary phase and median replacement operation on outliers in the abrupt change phase.

[0035] For example, industrial operating environments may experience electromagnetic harmonic interference and variations in sensor pin contact impedance, leading to discrete pulses and numerical distortions in the sampled data sequence. The data acquisition and preprocessing module constructs a sliding observation window within the cache space. For discrete energy consumption sequences of a single physical dimension within the sliding observation window, a sorting algorithm is invoked to rearrange the data samples in ascending order. Based on the overall data length of the sequence, numerical parameters located at the lower quartile are extracted using index addressing and established as the lower quartiles, and numerical parameters located at the upper quartiles are extracted and established as the upper quartiles. The algebraic difference between the upper and lower quartiles is calculated to obtain the interquartile range. A preset interquartile range factor is read, and the product of the preset interquartile range factor and the interquartile range is calculated. The specific value range of the preset interquartile range factor is limited to between 1.5 and 3.0. In this embodiment, for the power medium acquisition channel where electromagnetic interference is frequent, the factor is preferably set to 3.0 to filter out extreme spike crosstalk; for the natural gas and water resource medium acquisition channels where the changes are relatively gradual, the factor is preferably set to 1.5 to ensure the sensitivity to capture slight fluctuations caused by minor mechanical jamming of the equipment.

[0036] The upper detection limit is established by adding the product to the upper quartile, and the lower detection limit is established by subtracting the product from the lower quartile. Each target value within the current sliding observation window is iterated through. When a target value is greater than the upper detection limit or less than the lower detection limit, the corresponding value is determined to be an abnormal value.

[0037] Understandably, the data acquisition and preprocessing module uses the mean-standard deviation method as auxiliary verification logic. It calculates the arithmetic mean and standard deviation of all valid sample data in the same dimension within the current sliding observation window. When the algebraic difference between the detected target value and the arithmetic mean exceeds a set number of standard deviation measures, it is also determined to be an outlier.

[0038] The data acquisition and preprocessing module performs mean replacement or median replacement operations on the identified outlier values. It extracts the adjacent normal sampling points forward and backward from the point where the outlier value occurred on the time axis. It then calculates the local variance characteristics of the sequence of adjacent normal sampling points.

[0039] The specific method for obtaining and determining the steady-state judgment threshold is as follows: extract sensor data from 100 consecutive sampling cycles of the corresponding controlled energy-consuming equipment under historical static no-load or constant rated load conditions, calculate the absolute value of its background operating variance, and multiply the absolute value of the background operating variance by a redundancy safety factor of 1.2 to 1.5 (preferably 1.2 in this embodiment). The resulting product is calibrated as the steady-state judgment threshold. This calibration logic enables the judgment threshold to adapt to the dynamic benchmark of the inherent background noise of different underlying physical sensors.

[0040] If the local variance characteristic is less than the steady-state determination threshold, the data acquisition and preprocessing module classifies the physical state as a stationary phase and assigns a mean replacement operation. The arithmetic mean parameter of the adjacent normal sampling points is calculated and overwritten into the physical memory address occupied by the abnormal value. If the local variance characteristic is greater than or equal to the steady-state determination threshold, the data acquisition and preprocessing module classifies the physical state as a sudden change phase and redirects the median replacement operation. After sorting the adjacent normal sampling point sequence by numerical value, the numerical parameter at the center position is extracted and overwritten into the physical memory address occupied by the abnormal value.

[0041] Next, the replaced multi-dimensional energy consumption time-series data is distributed and stored using a distributed storage and computing platform. The distributed file system module includes a master name node deployed in the main control room and multiple data storage nodes distributed across different subnets. The distributed file system module truncates and divides the continuous data stream into multiple fixed-length standard physical data blocks according to a set upper limit threshold for byte length. The master name node issues data block write routing instructions based on the remaining available disk space variables reported by each data storage node. The data storage nodes perform disk writes of the standard physical data blocks. The distributed file system module implements a data redundancy and disaster recovery mechanism, cloning and replicating each standard physical data block into multiple copies, and writing them across physical racks to independent backup data storage nodes using addressing instructions.

[0042] Furthermore, such as Figure 2 As shown, the dynamic synchronization processing of the multi-dimensional energy consumption time-series data in distributed storage, combined with the physical response lag characteristics of the controlled energy-consuming equipment, includes: S201: Extract the timestamp parameter of the multidimensional energy consumption time series data from the distributed storage, and combine the timestamp parameter with the calendar to convert the timestamp parameter into a high-dimensional one-hot encoded sparse matrix to obtain the time period features; S202: Determine the delay step based on the total length of the transmission pipeline of the underlying controlled medium of the controlled energy-consuming equipment and the medium flow velocity; S203: Construct the physical response hysteresis characteristics of the controlled energy-consuming equipment by combining the delay step size and the corresponding historical time point energy consumption value of the controlled energy-consuming equipment in the same dimension; S204: Perform reduction processing on the multidimensional energy consumption time series data by combining the device ID, time period characteristics and physical response lag characteristics of the controlled energy-consuming equipment.

[0043] For example, dynamic data synchronization is achieved by extracting time-period features and lag features through the distributed storage and computing platform. The master task allocation node of the mapping and reduction computing framework pushes the feature extraction computing instruction subtasks across the network to the corresponding data storage nodes. The local memory controller of the data storage node establishes a virtual sliding time window in the random access memory. During the mapping processing stage, the timestamp parameter of the data sample set is extracted. The calendar logic parsing step is executed to extract numerical feature variables indicating that the time node is located at an independent hour sequence position within the day, numerical feature variables indicating that the date of the time node is located at a position within the natural week sequence, and binary Boolean type feature variables indicating whether the current date is defined as a weekend or holiday. The computing process converts the time-dimensional numerical feature variables into a high-dimensional one-hot encoded sparse matrix through matrix projection rules, thereby constituting the time-period features. Using the latest acquisition time point of the current sliding time window as the global reference coordinate, the system searches along the historical trajectory segment along the time axis. To effectively overcome the problem of the delay shift step size relying on experience in feature engineering, this system physically binds this step size to the fluid dynamics spatial parameters. The specific derivation steps are as follows: Extract the scalar of the total actual physical pipeline length between the source valve and the corresponding end monitoring sensor of the underlying controlled medium (such as chilled water or natural gas). Read the average fluid propagation linear velocity parameter of the medium under the current pipeline operating pressure. ;Will Divide by The absolute time of the actual physical transmission hysteresis is obtained; finally, this absolute time is divided by the sampling frequency step time of the underlying analog-to-digital conversion circuit of the system, and the result is rounded up to obtain the set delay shift step size. This calibration logic ensures that the retraction scale of the feature extraction receptive field of the temporal neural network on the time axis accurately matches the actual hysteresis period of material and energy flow in physical space, reducing the probability of feature and target misalignment.

[0044] Extract energy consumption numerical variable data of the same dimension corresponding to historical time points with a set delay shift step, and concatenate them to construct the lag feature. In the reduction processing stage, the aggregation computing node performs aggregation and coordinate alignment operations on the input feature stream based on the hardware number of the controlled energy-consuming device, i.e., the device ID, and fuses the output multi-dimensional energy consumption time series data after synchronous processing.

[0045] In one embodiment, the step of determining abnormal operating condition data by using a clustering algorithm and Newton's iteration method on the multidimensional energy consumption time-series data after dynamic synchronization processing includes: S301: Cluster the multidimensional energy consumption time series data after dynamic synchronization processing, and determine candidate abnormal operating condition data by combining the clustering results and the minority class proportion threshold. The minority class proportion threshold is calculated based on the prior statistical probability benchmark of the average failure interval time of the controlled energy-consuming equipment and the expected hardware failure rate. S302: The candidate abnormal operating condition data is verified using the Newton-Raphson iteration method. If it is determined that the candidate abnormal operating condition data violates the physical manufacturing limit of the controlled energy-consuming equipment, then the candidate abnormal operating condition data is determined to be the abnormal operating condition data.

[0046] For example, multiple sample data rows are selected from the synchronized multidimensional energy consumption time-series data to construct a high-dimensional feature training sample set. Based on the set cluster number parameter, a predetermined number of initial cluster center coordinate vectors are allocated and established in the coordinate space of the high-dimensional feature training sample set. The scalar distance from each independent data sample feature vector to all initial cluster center coordinate vectors is calculated using the Euclidean distance formula, and the independent data sample feature vectors are assigned to the corresponding cluster subset with the smallest geometric distance measurement value. The arithmetic mean component of the data samples within each cluster subset is calculated and updated as the new cluster center coordinate vector. The distance allocation and update steps are repeated. When the difference in the sum of squared errors between two adjacent iterations is lower than a preset convergence tolerance constant, the clustering network model is considered to have reached convergence. The proportion of sample features contained within each cluster subset is statistically analyzed. When the proportion of samples absorbed by a specific cluster subset is lower than the minority class percentage threshold, and the relative vector distance parameter between the coordinate vector of the cluster center represented by the specific cluster subset and the center of the mainstream cluster is greater than a predetermined safe dispersion radius, the specific cluster subset is determined to exhibit an abnormal pattern (i.e., candidate abnormal operating condition data). The minority class percentage threshold is preferably limited to 1% to 3%, and is derived from the prior statistical probability benchmark of the mean time between failures (MTBF) and the expected hardware failure rate specified on the factory nameplate of controlled energy-consuming equipment (such as water pump mechanical bearings and gas solenoid valves). The predetermined safe dispersion radius is specifically established as follows: extract globally valid data samples within the current sliding time window, calculate the covariance matrix of the multidimensional feature space, and set the predetermined safe dispersion radius to be equal to three times the standard deviation of the Mahalanobis distance corresponding to this covariance distribution.

[0047] The abnormal behavior identification and verification unit, combined with Newton's iteration method logic, supplements and confirms the suspicious data feature values ​​within the abnormal mode, that is, verifies the candidate abnormal operating condition data to confirm whether it is truly abnormal. The abnormal behavior identification and verification unit is pre-deployed with a nonlinear differential equation mathematical model library reflecting the balance between energy flow and medium flow conservation. The scalar element matrix of the suspicious data feature values ​​is used as the initial input and substituted into the model parameters. The mapping function result from the nonlinear differential equation mathematical model library is output, and the corresponding derivative matrix gradient partial derivative numerical results are simultaneously obtained using automatic differential link calculation. The model input parameter variables in the current iteration environment are subtracted from the quotient obtained by performing a floating-point division operation between the mapping function result and the derivative matrix gradient partial derivative numerical results to generate new parameter variables for subsequent iteration environments.

[0048] Taking the monitoring of intelligent water supply pumps in the pipeline network as an example, the mapping function that reflects the balance between energy flow and medium flow rate conservation... Specifically, this is instantiated as the pipeline network fluid damping pressure drop equation: Among them, the input parameters The flow rate of the suspected pipeline section used for verification; The real physical pressure difference absolute value readings collected in real time by the differential pressure sensors at both ends of the pipeline network; This is an empirical constant for comprehensive local resistance, pre-calibrated based on the pipeline's physical materials, fluid density, and pipe diameter. In engineering implementation, this empirical constant... A dimension conversion coefficient is incorporated to absorb and smooth out the difference in physical dimensions between the square of the fluid flow rate and the absolute value of the pressure difference. Correspondingly, the numerical result of the gradient partial derivative of the derivative matrix is ​​the first derivative of the equation. .

[0049] After a set number of iterative approximation calculations, if the final iterative convergence of the theoretical derivation solution value violates the physical manufacturing extreme value limit of the controlled component, then it is verified that the equipment operating condition corresponding to the suspicious data feature value has a physical structural failure, and the abnormal pattern recognition is determined to be effective, that is, the abnormal operating condition data is identified.

[0050] In one embodiment, the adversarial network needs to be pre-built before it can be invoked. Building the adversarial network includes: S7: Construct a generator network, which uses a fully connected network architecture to generate simulated sequences of multidimensional energy consumption time series data; S8: Construct a discriminant prediction network with a hybrid architecture of convolutional layers and bidirectional long short-term memory layers. The discriminant prediction network is trained based on the multidimensional energy consumption time series data and the simulated sequence. The process of using a pre-constructed adversarial network to process multidimensional energy consumption time-series data that has undergone dynamic synchronization processing and includes the abnormal operating condition data, to obtain the authenticity judgment probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment, includes: S401: The multidimensional energy consumption time series data, which includes the abnormal operating condition data, after dynamic synchronization processing is performed using the discriminant prediction network in the pre-constructed adversarial network to obtain the authenticity discrimination probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment.

[0051] For example, the generative adversarial network (GAN) energy consumption prediction module, i.e., the adversarial network, includes a generator network and a discriminant prediction network. The generator network uses a fully connected network architecture to generate simulated sequences. The generator network receives a high-dimensional random noise array matrix tensor from a pseudo-random dynamic sequence generation module. The numerical elements inside the high-dimensional random noise array matrix tensor follow a normal distribution. After entering the generator network, the high-dimensional random noise array matrix tensor passes through multiple layers of fully connected hidden nodes. The hidden node layers perform matrix inner product and bias scalar addition operations. The data is then fed into the modified linear unit activation layer. Batch normalization components are deployed between the hidden layer nodes to scale and normalize the data. After mapping operations by the hyperbolic tangent continuous function module, the output parameter range is constrained and the generated simulated sequence is output.

[0052] The discriminant prediction network employs a hybrid architecture combining convolutional layers and bidirectional long short-term memory (LSTM) layers. The network receives the synchronously processed multidimensional energy consumption time-series data and the generated simulated sequence. Local temporal features are first extracted via the convolutional layers. Multiple convolutional kernels with different receptive fields are configured within a single-order discrete convolutional computation layer. The convolutional kernels perform a unidirectional sliding coverage operation along the time-series axis. Within the local matrix block covered by the sliding field of view, a local inner product dot product operation is performed, multiplying and summing corresponding nodes to generate a spatiotemporal feature map matrix reflecting the local frequency domain response. Downsampling and dimensionality reduction are then performed via a spatial pooling layer. The feature matrix stream retaining the extreme response points is fed in series to the bidirectional LSM layer. The bidirectional LSM layer includes a forward propagation computation link and a backward propagation computation link. Control gate operators at the forward propagation link nodes cyclically read feature segments in ascending time order; synchronously operating backward propagation link nodes cyclically read feature segments in reverse time order. An abstract hidden layer fusion feature matrix is ​​generated by the forgetting control logic module, the input update control logic module, and the output increment control logic module to extract long-term temporal dependencies.

[0053] Finally, the discriminant prediction network outputs the authenticity probability and the future energy consumption prediction value through its dual output endpoints. The classification branch, located at the beginning of the dual output endpoints, performs fully connected dimensionality reduction on the abstract hidden layer fusion feature matrix. At the terminal node, it performs numerical mapping using a logistic transformation function, outputting a real scalar between zero and its upper limit; this real scalar represents the authenticity probability. The regression branch, located at the end of the dual output endpoints, performs fully connected linear mapping and nonlinear activation on the input tensor, outputting the future energy consumption prediction value with power physical dimensions. During training, an alternating optimization control logic is used to construct a joint loss function including a cross-entropy loss operator and a mean squared error operator, and the network connection weights are updated using a backpropagation algorithm.

[0054] Furthermore, given that the probability of authenticity discrimination (dimensionless, domain between 0 and 1) and the predicted future energy consumption (real numbers with large numerical values ​​in kilowatts or cubic meters) are highly susceptible to backpropagation gradient flooding due to significant differences in their numerical magnitudes when performing joint calculations, the joint loss function (denoted as Ltotal) incorporates a nonlinear alignment operator for the physical magnitudes, the algebraic expression of which is:

[0055] In the formula, For cross-entropy loss operator; It is a mean square error operator with physical dimensions attached; This represents the total installed rated power of the currently controlled units or the maximum range constant of the pipeline network. This formula introduces a dimensionless ratio within the logarithmic term. From a rigorous mathematical perspective, the interference of physical dimensions was eliminated, and then the natural logarithm function was used. The massive physical squared error is nonlinearly reduced in size to align with the logarithmic fractional scale domain that matches the probability distribution. For constant proportional weights, for example: ; To dynamically adjust the weight multiplier. The dynamically adjusted weight multiplier... The method to obtain it is: set an initial value. The value is between 0.5 and 1.0, and it is calculated according to a preset attenuation formula. With training iteration rounds Decreasing, of which The natural constant and the decay constant are the constants. The optimal value is 0.01. The introduction of this mathematical structure to ensure smooth convergence of the model effectively improves the smooth convergence stability of the dual-output adversarial network under heterogeneous physical dimensions.

[0056] In one embodiment, the multidimensional energy consumption time-series data includes natural gas consumption, water resource consumption, and electricity consumption dimensions; The determination of target control parameters for regulating the controlled energy-consuming equipment based on the predicted future energy consumption includes: S501: Construct a comprehensive objective function that covers the aforementioned natural gas consumption dimension, water resource consumption dimension, and electricity consumption dimension; S502: Solve for the target control parameters using a convex optimization solver with the goal of minimizing the value of the integrated objective function.

[0057] For example, in this embodiment, the robust optimization and control module incorporates a genetic algorithm optimization engine and a convex optimization solver to execute the initialization modeling construction instructions for the comprehensive objective function. Specifically, the multidimensional energy consumption time-series data covers the dimensions of natural gas consumption, water resource consumption, and electricity consumption. A comprehensive objective function covering these dimensions is constructed. The value of the comprehensive objective function is equal to the sum of the output values ​​of the natural gas consumption model, the water resource consumption model, and the electricity consumption model. The output value of the natural gas consumption model is obtained by performing a product arithmetic operation on the predicted real-time natural gas flow reading, the statistical time step, and the gas efficiency coefficient scalar. The output value of the water resource consumption model is obtained by performing a product arithmetic operation on the predicted real-time water flow reading, the statistical time step, and the water conveyance efficiency coefficient scalar. The output value of the electricity consumption model is obtained by performing a product arithmetic operation on the predicted real-time operating power reading, the statistical time step, and the power efficiency coefficient scalar.

[0058] It is particularly important to note that, since natural gas consumption (volume dimension), water resource consumption (mass / volume dimension), and electricity consumption (electrical work dimension) belong to different physical dimensions, before the final summation calculation of the comprehensive objective function, the gas efficiency coefficient, water transmission efficiency coefficient, and electrical energy efficiency coefficient are internally integrated with a conversion operator that unifies the dimensions. Specifically, these three efficiency coefficients not only include the mechanical work conversion rate specified on the nameplate of the controlled energy-consuming equipment (values ​​ranging from 0.6 to 0.98), but also include industry standard conversion factors that uniformly map and convert the corresponding medium consumption into standard coal consumption equivalent (unit: kg standard coal / kgce) or comprehensive carbon emission equivalent. Through this multi-dimensional matrix multiplication transformation, the energy consumption values ​​of the three heterogeneous physical media are standardized, cleaned, and uniformly mapped into equivalent scalars with the same environmental and economic dimensions, thereby ensuring the dual feasibility of the multi-dimensional objective comprehensive function summation calculation in terms of physical laws and mathematical logic.

[0059] The robust optimization and control module utilizes the convex optimization solver to minimize the comprehensive objective function. It performs numerical optimization within the convex domain of feasible solutions defined by the physical limit constraint calibration matrix, locating the floating-point parameters of the pre-optimization decision sequence. Specifically, the convex optimization solver employs an interior-point algorithm or directly calls mature convex optimization function libraries such as CVXPY.

[0060] The genetic algorithm optimization engine converts the floating-point parameters of the pre-optimization decision sequence into a combination of low-level proportional-integral-differential control parameter matrices. Before optimization, the specific control parameters of the genetic algorithm are preset: the initial population size is set to 50 to 200 (preferably 100 in this embodiment), the crossover operation probability is set to 0.7 to 0.9 (preferably 0.8), the numerical positional mutation operation probability is set to 0.01 to 0.1 (preferably 0.05), and the maximum generation limit parameter is set to 100 to 500 generations. The genetic algorithm optimization engine generates the initial population floating-point matrix and performs crossover recombination operations and numerical positional mutation operations. The array of new generation chromosome entity units is brought into the controlled object simulation environment to calculate the cumulative area index parameter of the integral error of the follower response.

[0061] Specifically, the cumulative area index parameter of the servo response integral error is not a simple algebraic geometric area summation, but rather deeply integrates the time-in-absolute-error integral criterion (ITAE criterion) from industrial automatic control theory. Its evaluation definite integral analytical expression is as follows: In the formula, The cumulative area index parameter for the integral error of the follow-up response; The set duration for a single control simulation tracking observation; The physical evolution time is the independent variable calculated from the moment the simulation command is issued; To control the expected trajectory to the target; This represents the actual tracking response trajectory at the underlying level of the controlled device. A time variable is introduced into the integral term of this absolute error. As a dynamically increasing penalty multiplier, its function is to allow for large transient errors caused by physical hysteresis in the early stages of operation of heavy machinery, and to utilize the linearly increasing penalty multiplier over time. t Increasing numerical penalty weights are applied to the high-frequency mechanical jitter and steady-state residuals remaining in the later stages of the operation, thereby prompting the underlying genetic algorithm to select the combination of control parameters that can effectively suppress the physical oscillations of the underlying equipment.

[0062] A fitness evaluation scale is established by performing a reciprocal algebraic mapping operation. A roulette wheel probability selection and elimination algorithm is then initiated to eliminate low-fitness arrays. After iteratively reaching the highest limiting algebraic parameter, the optimal floating-point vector sequence feature combination with the highest fitness ranking is extracted and established as the target control parameter for the controlled energy-consuming equipment.

[0063] In one embodiment, combining the physical response hysteresis characteristics and the authenticity discrimination probability, the target control parameters are asynchronously sent to different types of controlled energy-consuming devices, including: S601: Based on the physical response hysteresis characteristics, the controlled energy-consuming equipment is divided into controlled energy-consuming equipment of slow response type and controlled energy-consuming equipment of fast response type; S602: The target control parameters are preferentially sent to the controlled energy-consuming equipment of the slow response type to realize equipment regulation; S603: Monitor the actual physical flow of the pipeline network where the controlled energy-consuming equipment is located, and when the actual physical flow exceeds a preset threshold, send the target control parameters to the controlled energy-consuming equipment of the fast response type, and regulate the controlled energy-consuming equipment of the fast response type in combination with the actual discrimination probability.

[0064] For example, the controlled energy-consuming equipment in this embodiment includes water control equipment, gas control equipment, and electric actuators. The robustness optimization and control module retrieves the timestamp parameter records of each control command issuance from the historical spatiotemporal database, as well as the timestamp parameter records of the state taking effect corresponding to the physical sensor feedback parameters crossing the steady-state dead zone boundary.

[0065] When reading and extracting historical records, the specific capture and determination logic for the state effective timestamp corresponding to crossing the steady-state dead zone boundary is a first-order differential threshold determination method: the underlying storage controller performs a first-order difference operation on the time-series discrete measurement values ​​returned by physical sensors (such as pipeline flow meters or servo motor current transformers) at adjacent time nodes to obtain the instantaneous rate of change. Only when the absolute value of this instantaneous rate of change exceeds the preset hardware background noise fluctuation extreme value (e.g., the fluid rate of change is continuously greater than...) within three consecutive communication sampling periods... Only after this process is complete can the system confirm that the controlled mechanical component has overcome the static friction dead zone and generated substantial macroscopic physical displacement. The specific method for determining the extreme value of the hardware background noise fluctuation is as follows: extract data records from the corresponding physical sensor for 100 consecutive communication sampling cycles under the steady-state condition of the equipment shutdown, calculate the maximum absolute value of the instantaneous change rate of the first-order difference between adjacent nodes, and multiply this maximum absolute value by a redundancy safety factor of 1.2 to 1.5 (preferably 1.2) for calibration. At this time, the system will extract the microsecond-level hardware clock point corresponding to the first threshold breakthrough moment and solidify it into the register as the state activation timestamp. This microelectromechanical differential locking logic effectively filters out the illusory actions caused by initial electromagnetic jitter and the water hammer effect at the start of fluid flow, effectively improving the accuracy of response hysteresis attribute extraction.

[0066] The clock delay difference data scalar between the control command issuance timestamp parameter record and the status activation timestamp parameter record is calculated. This clock delay difference data scalar is defined as the time hysteresis parameter corresponding to the controlled energy-consuming equipment, i.e., the physical response hysteresis attribute. A descending sorting algorithm is performed on the time hysteresis parameters of all controlled energy-consuming equipment. The robust optimization and control module groups the water conservancy control equipment with the largest time hysteresis parameter ranking and the gas control equipment with the middle ranking into the slow response equipment group. The power execution equipment with the smallest time hysteresis parameter ranking is grouped into the fast response equipment group.

[0067] Furthermore, in this embodiment, the target control parameters include proportional-integral-derivative control parameters; The controlled energy-consuming devices of the fast-response type are regulated based on the actual discrimination probability, including: S604: Determine the dynamic smoothing weight of the controlled energy-consuming equipment of the fast response type based on the authenticity discrimination probability; S605: The control slope of the controlled energy-consuming equipment of the fast response type is constrained by the dynamic smoothing weight as the attenuation multiplier, and the control slope is determined based on the integral time constant in the proportional-integral-derivative control parameters.

[0068] For example, the robustness optimization and control module creates an asynchronous execution sequence queuing list topology within the bus cache space, and based on this, prioritizes sending the associated target control parameters to the slow-response device group. Specifically, when constructing the asynchronous execution sequence, the robustness optimization and control module arranges the water conservancy control device at the first execution order, the gas control device in the middle execution order, and the power execution device at the last execution order. During the instruction issuance phase, the underlying communication module prioritizes communication with the water conservancy control device at the first execution order, sending the corresponding target control parameters. A physical state handshake monitoring loop logic resident in memory is simultaneously initiated. This physical state handshake monitoring loop logic is used to continuously read the actual physical flow feedback parameters of the pipeline network. During this period, the transmission channel to devices in subsequent orders is blocked.

[0069] The derivation logic for obtaining the preset confirmation threshold is as follows: Extract the steady-state expected flow rate value from the target control parameters first issued to the first-order hydraulic control equipment; read the preset dead zone tolerance ratio constant (preferably calibrated to 10% to 15% based on mechanical transmission characteristics) used to characterize overcoming the static friction of mechanical valves and the resistance of starting water hammer; multiply the steady-state expected flow rate value by the dead zone tolerance ratio constant to obtain the offset increment value; and add this offset increment value to the initial flow background value of the pipeline network at the time the command is issued; finally, sum them to obtain the preset confirmation threshold for this round. This physical state handshake mechanism ensures from a judgment logic perspective that: only when monitoring confirms that the slow-speed mechanical equipment has truly generated a substantial macroscopic fluid displacement sufficient to break through the static dead zone will the signal be released to the fast-response equipment. From a physical execution level, this effectively avoids fluid idling and dry burning or electrical transient energy accumulation and oscillation caused by asynchronous start-up of heterogeneous equipment.

[0070] The unlock command is only released after the actual physical flow feedback of the pipeline network exceeds a preset confirmation threshold. The unlocking channels for the gas control equipment and the electrical actuator are then sequentially withdrawn.

[0071] Furthermore, the target control parameters include proportional-integral-derivative (PID) control parameters. The control slope of the fast-response device group is determined by the integral time constant within the PID control parameters. The robust optimization and control module extracts the true discrimination probability floating-point value output by the discriminant prediction network. The true discrimination probability floating-point value is converted into the dynamic smoothing weights using the reciprocal nonlinear proportional mapping constraint operator module.

[0072] Specifically, the reciprocal nonlinear proportional mapping constraint operator module internally encapsulates an algebraic transformation relation, the calculation derivation formula of which is as follows: ; in, The dynamic smoothing weights representing the output of the operation; The probability of authenticity determination is a floating-point value, and its domain is limited to . ; The basic response retains a constant, with a fixed value. This is used to ensure the system's basic control follow-up capability under high prediction confidence. The penalty mapping gain coefficient is preset to address the physical hysteresis characteristics of different devices, and its value range is set within [range missing]. to between; To place it in the denominator term, to prevent the microprocessor from... The minimum positive bias constant that causes numerical overflow or abnormal system termination when division is performed near zero is fixed in the system. .

[0073] The industrial control mathematical logic of this nonlinear mapping operator lies in: when the discrimination and prediction network detects that the current front-end energy consumption sequence contains drastic changes in operating conditions or strong unknown abnormal disturbances (i.e., manifested as true probability) When the value is extremely low and approaches 0, the dynamic smoothing weights output are based on this reciprocal fraction structure. It will be safely and nonlinearly amplified to an order of magnitude much greater than 1 along an inverse proportional curve, thereby triggering a strong damping suppression mechanism at the underlying level through parameter passing.

[0074] Upon detecting that the slow-response device group has received an instruction and is in the physical action transition phase, the robust optimization and control module locks the integral time constant allocated to the fast-response device group. When executing the constraint operation, the dynamic smoothing weight is applied as a decay multiplier to the integral time constant. The integral time constant is amplified by the dynamic smoothing weight, causing it to increase accordingly.

[0075] The algebraic execution relation for this constraint operation within the underlying control unit is: .in, The initial integral time constant (in seconds) calculated for the fast-response device to comprehensively optimize the solver. This refers to the updated integral time constant actually programmed into the underlying hardware PID controller register after constraint amplification. In the classic proportional-integral-derivative (PID) automatic control logic, the integral control force that eliminates steady-state error is essentially inversely proportional to the integral time constant (as a denominator term in the formula). This is because the aforementioned mapping formula ensures dynamic smoothing of the weights. This scalar multiplication operation directly amplifies the integral time constant proportionally from a mathematical perspective, thereby substantially reducing the cumulative integral step size of the control chip within a single sampling period. This flexible approach, achieved solely through dynamic resetting of software algorithm parameters, smooths out the peaks and tails the original step-type power output curve of pure electrical equipment without requiring additional physical or mechanical damping components. This forces the slope of the electrical signal's ramp-up to slow down to the tens of seconds level, thus achieving deep physical resonance with the slow rotation of the mechanical water valve on the time axis.

[0076] This numerical amplification process reduces the sensitivity of the fast-response device group to control errors, thereby smoothing out the steepness of output power changes and exhibiting a gradual, ramp-like motion that matches the mechanical action state. The constraint operation forces the motion variation curve of the fast-response device group to fit the hysteresis motion curve of the slow-response device group.

[0077] Furthermore, regulating the controlled energy-consuming equipment includes: S606: For different controlled energy-consuming devices, the target control parameters are mapped to different actual execution variables, and the actual execution variables are regulated; Specifically, for HVAC equipment among the controlled energy-consuming devices of the aforementioned fast-response type, the actual execution variable is the compressor operating frequency or the fan speed; For the energy storage battery charge / discharge controller in the controlled energy-consuming device of the aforementioned fast-response type, the actual execution variable is the charge / discharge cutoff voltage or the charge / discharge rate limit; For photovoltaic inverters in controlled energy-consuming devices of the aforementioned fast-response type, the actual execution variable is the maximum power point tracking current reference value; For the intelligent water supply pump and gas-fired wall-mounted boiler in the controlled energy-consuming equipment of the slow response type, the actual execution variables are the opening degree of the water valve and the opening degree of the gas nozzle.

[0078] like Figure 3 As shown, another embodiment of the present invention also provides a distributed energy consumption optimization system based on multimodal data synchronization and generation, comprising: The data acquisition module is used to acquire multi-dimensional energy consumption time-series data of controlled energy-consuming devices in the distributed integrated energy management and control network, and to process outliers in the multi-dimensional energy consumption time-series data. The first processing module is used to perform distributed storage of the processed multidimensional energy consumption time series data, and to perform dynamic synchronization processing of the distributed storage of the multidimensional energy consumption time series data in combination with the physical response lag characteristics of the controlled energy-consuming equipment. The first determining module is used to determine abnormal operating condition data by using clustering algorithm and Newton's iteration method on the multidimensional energy consumption time series data after dynamic synchronization processing. The second processing module is used to process the multidimensional energy consumption time series data containing the abnormal operating condition data after dynamic synchronization processing using a pre-built adversarial network, and to obtain the authenticity judgment probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment. The second determining module is used to determine the target control parameters for regulating the controlled energy-consuming equipment based on the future energy consumption prediction value. The control module is used to combine the physical response hysteresis characteristics and the authenticity discrimination probability to asynchronously send the target control parameters to different types of controlled energy-consuming devices, so as to control the controlled energy-consuming devices and fit the action variation curve of the fast-response type controlled energy-consuming device with the hysteresis action curve of the slow-response type controlled energy-consuming device.

[0079] In one embodiment, the outlier processing of the multidimensional energy consumption time-series data includes: The sliding window method and the interquartile range method are used to detect outliers in the multidimensional energy consumption time series data corresponding to different sliding windows. The interquartile range multiplier is set differently for different types of acquisition channels. Calculate local variance based on sampled data before and after outliers; Based on the relationship between local variance and a preset steady-state threshold, it is determined whether the controlled energy-consuming equipment is in a stable or abrupt stage. The outliers in the stationary phase are replaced by the mean, and the outliers in the abrupt change phase are replaced by the median.

[0080] In one embodiment, the dynamic synchronization processing of the multi-dimensional energy consumption time-series data in distributed storage combined with the physical response lag characteristics of the controlled energy-consuming equipment includes: Extract the timestamp parameter from the multidimensional energy consumption time series data in the distributed storage, and combine the timestamp parameter with the calendar to convert the timestamp parameter into a high-dimensional one-hot encoded sparse matrix to obtain the time period features; The delay step is determined based on the total length of the transmission pipeline of the underlying controlled medium of the controlled energy-consuming equipment and the medium flow velocity. The physical response hysteresis characteristics of the controlled energy-consuming equipment are constructed by combining the delay step size and the corresponding energy consumption value of the controlled energy-consuming equipment at the same historical time point. The multidimensional energy consumption time series data is reduced by combining the device ID, time period characteristics, and physical response lag characteristics of the controlled energy-consuming equipment.

[0081] In one embodiment, the step of determining abnormal operating condition data by using a clustering algorithm and Newton's iteration method on the multidimensional energy consumption time-series data after dynamic synchronization processing includes: Clustering is performed on the multidimensional energy consumption time series data after dynamic synchronization processing, and candidate abnormal operating condition data are determined by combining the clustering results and the minority class proportion threshold. The minority class proportion threshold is calculated based on the prior statistical probability benchmark of the average failure interval time of the controlled energy-consuming equipment and the expected hardware failure rate. The candidate abnormal operating condition data is verified using the Newton-Raphson iteration method. If it is determined that the candidate abnormal operating condition data violates the physical manufacturing limit of the controlled energy-consuming equipment, then the candidate abnormal operating condition data is determined to be the abnormal operating condition data.

[0082] In one embodiment, constructing the adversarial network includes: A generator network is constructed, which uses a fully connected network architecture to generate simulated sequences of multidimensional energy consumption time-series data. A discriminant prediction network is constructed using a hybrid architecture of convolutional layers and bidirectional long short-term memory layers. The discriminant prediction network is trained based on the multidimensional energy consumption time-series data and simulated sequences. The process of using a pre-constructed adversarial network to process multidimensional energy consumption time-series data that has undergone dynamic synchronization processing and includes the abnormal operating condition data, to obtain the authenticity judgment probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment, includes: The discriminant prediction network in the pre-constructed adversarial network is used to process the multidimensional energy consumption time series data that has been dynamically synchronized and includes the abnormal operating condition data, so as to obtain the authenticity discrimination probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment.

[0083] In one embodiment, the multidimensional energy consumption time-series data includes natural gas consumption, water resource consumption, and electricity consumption dimensions; The determination of target control parameters for regulating the controlled energy-consuming equipment based on the predicted future energy consumption includes: Construct a comprehensive objective function that covers the aforementioned dimensions of natural gas consumption, water resource consumption, and electricity consumption; The target control parameters are solved by using a convex optimization solver to minimize the value of the integrated objective function.

[0084] In one embodiment, combining the physical response hysteresis characteristics and the authenticity discrimination probability, the target control parameters are asynchronously sent to different types of controlled energy-consuming devices, including: Based on the physical response hysteresis characteristics, the controlled energy-consuming equipment is divided into slow-response type controlled energy-consuming equipment and fast-response type controlled energy-consuming equipment; The target control parameters are preferentially sent to slow-response controlled energy-consuming devices to achieve device regulation; The actual physical flow of the pipeline network where the controlled energy-consuming equipment is located is monitored, and when the actual physical flow exceeds a preset threshold, the target control parameters are sent to the controlled energy-consuming equipment of the fast response type, and the controlled energy-consuming equipment of the fast response type is regulated in combination with the actual discrimination probability.

[0085] In one embodiment, the target control parameters include proportional-integral-derivative control parameters; The controlled energy-consuming devices of the fast-response type are regulated based on the actual discrimination probability, including: The dynamic smoothing weights of the controlled energy-consuming devices of the fast response type are determined based on the authenticity discrimination probability. The dynamic smoothing weight is used as an attenuation multiplier to constrain the control slope of the fast-response type controlled energy-consuming equipment, and the control slope is determined based on the integral time constant in the proportional-integral-derivative control parameters.

[0086] In one embodiment, regulating the controlled energy-consuming device includes: For different controlled energy-consuming devices, the target control parameters are mapped to different actual execution variables, and the actual execution variables are regulated. Specifically, for HVAC equipment among the controlled energy-consuming devices of the aforementioned fast-response type, the actual execution variable is the compressor operating frequency or the fan speed; For the energy storage battery charge / discharge controller in the controlled energy-consuming device of the aforementioned fast-response type, the actual execution variable is the charge / discharge cutoff voltage or the charge / discharge rate limit; For photovoltaic inverters in controlled energy-consuming devices of the aforementioned fast-response type, the actual execution variable is the maximum power point tracking current reference value; For the intelligent water supply pump and gas-fired wall-mounted boiler in the controlled energy-consuming equipment of the slow response type, the actual execution variables are the opening degree of the water valve and the opening degree of the gas nozzle.

[0087] Another embodiment of the present invention also provides an electronic device, comprising: One or more processors; Memory, configured to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the distributed energy consumption optimization method based on multimodal data synchronization and generation as described above.

[0088] Furthermore, one embodiment of the present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the distributed energy consumption optimization method based on multimodal data synchronization and generation as described above. It should be understood that the various solutions in this embodiment have the corresponding technical effects in the above-described method embodiments, and will not be repeated here.

[0089] Furthermore, embodiments of the present invention also provide a computer program product, which is tangibly stored on a computer-readable medium and includes computer-readable instructions that, when executed, cause at least one processor to perform a distributed energy consumption optimization method based on multimodal data synchronization and generation, as described in the embodiments above.

[0090] It should be noted that the computer storage medium of the present invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access storage medium (RAM), a read-only storage medium (ROM), an erasable programmable read-only storage medium (EPROM or flash memory), an optical fiber, a portable compact disk read-only storage medium (CD-ROM), an optical storage medium, a magnetic storage medium, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. In the present invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program configured for use by or in connection with an instruction execution system, system, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, antenna, optical fiber, RF, etc., or any suitable combination thereof.

[0091] Furthermore, those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0092] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0093] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0094] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

Claims

1. A distributed energy consumption optimization method based on multimodal data synchronization and generation, characterized in that, include: Collect multidimensional energy consumption time-series data of controlled energy-consuming devices in a distributed integrated energy management and control network, and perform outlier processing on the multidimensional energy consumption time-series data; The processed multidimensional energy consumption time series data is distributed and stored in a distributed manner. The distributed stored multidimensional energy consumption time series data is then dynamically synchronized in combination with the physical response lag characteristics of the controlled energy-consuming equipment. Clustering algorithms and Newton's iteration method are used to determine abnormal operating condition data for the multidimensional energy consumption time series data after dynamic synchronization processing; By using a pre-built adversarial network to process the multidimensional energy consumption time series data that has been dynamically synchronized and includes the abnormal operating condition data, the authenticity judgment probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment are obtained. Based on the predicted future energy consumption values, target control parameters for regulating the controlled energy-consuming equipment are determined. Combining the physical response hysteresis characteristics and the authenticity discrimination probability, the target control parameters are asynchronously sent to different types of controlled energy-consuming devices to regulate the controlled energy-consuming devices so that the action variation curve of the fast-response type controlled energy-consuming device is fitted with the hysteresis action curve of the slow-response type controlled energy-consuming device.

2. The distributed energy consumption optimization method based on multimodal data synchronization and generation according to claim 1, characterized in that, The outlier processing of the multidimensional energy consumption time-series data includes: The sliding window method and the interquartile range method are used to detect outliers in the multidimensional energy consumption time series data corresponding to different sliding windows. The interquartile range multiplier is set differently for different types of acquisition channels. Calculate local variance based on sampled data before and after outliers; Based on the relationship between local variance and a preset steady-state threshold, it is determined whether the controlled energy-consuming equipment is in a stable or abrupt stage. The outliers in the stationary phase are replaced by the mean, and the outliers in the abrupt change phase are replaced by the median.

3. The distributed energy consumption optimization method based on multimodal data synchronization and generation according to claim 1, characterized in that, The multi-dimensional energy consumption time-series data from the distributed storage is dynamically synchronized in conjunction with the physical response lag characteristics of the controlled energy-consuming equipment, including: Extract the timestamp parameter from the multidimensional energy consumption time series data in the distributed storage, and combine the timestamp parameter with the calendar to convert the timestamp parameter into a high-dimensional one-hot encoded sparse matrix to obtain the time period features; The delay step is determined based on the total length of the transmission pipeline of the underlying controlled medium of the controlled energy-consuming equipment and the medium flow velocity. The physical response hysteresis characteristics of the controlled energy-consuming equipment are constructed by combining the delay step size and the corresponding energy consumption value of the controlled energy-consuming equipment at the same historical time point. The multidimensional energy consumption time series data is reduced by combining the device ID, time period characteristics, and physical response lag characteristics of the controlled energy-consuming equipment.

4. The distributed energy consumption optimization method based on multimodal data synchronization and generation according to claim 1, characterized in that, The process of determining abnormal operating condition data by using clustering algorithms and Newton's iteration method on the multidimensional energy consumption time-series data after dynamic synchronization processing includes: Clustering is performed on the multidimensional energy consumption time series data after dynamic synchronization processing, and candidate abnormal operating condition data are determined by combining the clustering results and the minority class proportion threshold. The minority class proportion threshold is calculated based on the prior statistical probability benchmark of the average failure interval time of the controlled energy-consuming equipment and the expected hardware failure rate. The candidate abnormal operating condition data is verified using the Newton-Raphson iteration method. If it is determined that the candidate abnormal operating condition data violates the physical manufacturing limit of the controlled energy-consuming equipment, then the candidate abnormal operating condition data is determined to be the abnormal operating condition data.

5. The distributed energy consumption optimization method based on multimodal data synchronization and generation according to claim 1, characterized in that, Constructing the adversarial network includes: A generator network is constructed, which uses a fully connected network architecture to generate simulated sequences of multidimensional energy consumption time-series data. A discriminant prediction network is constructed using a hybrid architecture of convolutional layers and bidirectional long short-term memory layers. The discriminant prediction network is trained based on the multidimensional energy consumption time-series data and simulated sequences. The process of using a pre-constructed adversarial network to process multidimensional energy consumption time-series data that has undergone dynamic synchronization processing and includes the abnormal operating condition data, to obtain the authenticity judgment probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment, includes: The discriminant prediction network in the pre-constructed adversarial network is used to process the multidimensional energy consumption time series data that has been dynamically synchronized and includes the abnormal operating condition data, so as to obtain the authenticity discrimination probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment.

6. The distributed energy consumption optimization method based on multimodal data synchronization and generation according to claim 1, characterized in that, The multidimensional energy consumption time series data includes natural gas consumption, water resource consumption, and electricity consumption dimensions; The determination of target control parameters for regulating the controlled energy-consuming equipment based on the predicted future energy consumption includes: Construct a comprehensive objective function that covers the aforementioned dimensions of natural gas consumption, water resource consumption, and electricity consumption; The target control parameters are solved by using a convex optimization solver to minimize the value of the integrated objective function.

7. The distributed energy consumption optimization method based on multimodal data synchronization and generation according to claim 1, characterized in that, Combining the physical response hysteresis characteristics and the authenticity discrimination probability, the target control parameters are asynchronously sent to different types of controlled energy-consuming devices, including: Based on the physical response hysteresis characteristics, the controlled energy-consuming equipment is divided into slow-response type controlled energy-consuming equipment and fast-response type controlled energy-consuming equipment; The target control parameters are preferentially sent to slow-response controlled energy-consuming devices to achieve device regulation; The actual physical flow of the pipeline network where the controlled energy-consuming equipment is located is monitored, and when the actual physical flow exceeds a preset threshold, the target control parameters are sent to the controlled energy-consuming equipment of the fast response type, and the controlled energy-consuming equipment of the fast response type is regulated in combination with the actual discrimination probability.

8. The distributed energy consumption optimization method based on multimodal data synchronization and generation according to claim 7, characterized in that, The target control parameters include proportional-integral-derivative control parameters; The controlled energy-consuming devices of the fast-response type are regulated based on the actual discrimination probability, including: The dynamic smoothing weights of the controlled energy-consuming devices of the fast response type are determined based on the authenticity discrimination probability. The dynamic smoothing weight is used as an attenuation multiplier to constrain the control slope of the fast-response type controlled energy-consuming equipment, and the control slope is determined based on the integral time constant in the proportional-integral-derivative control parameters.

9. The distributed energy consumption optimization method based on multimodal data synchronization and generation according to claim 1, characterized in that, Regulating the controlled energy-consuming equipment includes: For different controlled energy-consuming devices, the target control parameters are mapped to different actual execution variables, and the actual execution variables are regulated. Specifically, for HVAC equipment among the controlled energy-consuming devices of the aforementioned fast-response type, the actual execution variable is the compressor operating frequency or the fan speed; For the energy storage battery charge / discharge controller in the controlled energy-consuming device of the aforementioned fast-response type, the actual execution variable is the charge / discharge cutoff voltage or the charge / discharge rate limit; For photovoltaic inverters in controlled energy-consuming devices of the aforementioned fast-response type, the actual execution variable is the maximum power point tracking current reference value; For the intelligent water supply pump and gas-fired wall-mounted boiler in the controlled energy-consuming equipment of the slow response type, the actual execution variables are the opening degree of the water valve and the opening degree of the gas nozzle.

10. A distributed energy consumption optimization system based on multimodal data synchronization and generation, characterized in that, include: The data acquisition module is used to acquire multi-dimensional energy consumption time-series data of controlled energy-consuming devices in the distributed integrated energy management and control network, and to process outliers in the multi-dimensional energy consumption time-series data. The first processing module is used to perform distributed storage of the processed multidimensional energy consumption time series data, and to perform dynamic synchronization processing of the distributed storage of the multidimensional energy consumption time series data in combination with the physical response lag characteristics of the controlled energy-consuming equipment. The first determining module is used to determine abnormal operating condition data by using clustering algorithm and Newton's iteration method on the multidimensional energy consumption time series data after dynamic synchronization processing. The second processing module is used to process the multidimensional energy consumption time series data containing the abnormal operating condition data after dynamic synchronization processing using a pre-built adversarial network, and to obtain the authenticity judgment probability of the data and the future energy consumption prediction value of the controlled energy-consuming equipment. The second determining module is used to determine the target control parameters for regulating the controlled energy-consuming equipment based on the future energy consumption prediction value. The control module is used to combine the physical response hysteresis characteristics and the authenticity discrimination probability to asynchronously send the target control parameters to different types of controlled energy-consuming devices, so as to control the controlled energy-consuming devices and fit the action variation curve of the fast-response type controlled energy-consuming device with the hysteresis action curve of the slow-response type controlled energy-consuming device.

Images