Demand side response-oriented multi-source heterogeneous data fusion method and system

Abstract

The invention discloses a demand side response-oriented multi-source heterogeneous data fusion method and system, and the method comprises the steps: obtaining multi-source heterogeneous data from a plant station collection device, the data comprising high-frequency electric energy data, device parameter data and DSR business index data; accessing a peak-valley-flat price signal of a power grid dispatching center and a DR event starting notification, identifying a current DSR scene type, and setting a differentiated fusion target; constructing an uplink aggregation + downlink prediction dual-channel alignment model, and based on the fusion target, performing dimension reduction alignment on the high-frequency electric energy data to generate aligned high-frequency electric energy data, and performing dimension raising alignment on the DSR service index data to generate aligned DSR service index data; and according to the current DSR scene type, fusing the aligned high-frequency electric energy data, the device parameter data and the aligned DSR service index data to obtain a fused comprehensive data set.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and more specifically, to a method and system for multi-source heterogeneous data fusion oriented towards demand-side response. Background Technology

[0002] With the advancement of new power system construction and the continuous increase in renewable energy penetration, the power grid supply and demand balance faces greater volatility challenges. Demand-side response, as an important means of flexible resource regulation, has been widely used in scenarios such as peak shaving and valley filling, frequency regulation, and reserve support. Its large-scale implementation highly depends on the accurate, real-time, and multi-dimensional data support provided by centralized data acquisition devices at power plants and substations.

[0003] Current mainstream data acquisition devices are capable of acquiring voltage and current waveforms at the 0.02-second level, equipment status parameters at the minute level (such as CT / PT ratio error and sampling frequency offset), and DSR aggregated indicators at the 15-minute to hour level (such as response capacity, participation rate, and compliance rate), making them a veritable "data bridge" connecting the power grid and users. However, in practical applications, these data exhibit significant multi-source heterogeneity:

[0004] (1) Large difference in time scale: The update cycle of high-frequency data is 20ms, while the decision cycle of DSR is more than 15 minutes, and the time resolution of the two differs by hundreds of times.

[0005] (2) Diverse data types: including analog quantities (voltage and current), digital quantities (switching status), and statistical quantities (average power and error rate);

[0006] (3) Conflicting business objectives: During peak hours, rapid response is required, emphasizing low data latency; during off-peak hours, fair settlement is required, emphasizing high data accuracy and traceability.

[0007] Existing data fusion methods mostly rely on traditional weighted averaging or Kalman filtering, failing to fully consider the differentiated needs of DSR scenarios. For example, during peak load periods, high-frequency data fusion delays exceeding 50ms can cause control command lag, resulting in actual response volumes being more than 10% lower than expected. Conversely, during off-peak billing periods, if a 0.8% drift in the ratio error of a transformer in a certain distribution area is not detected in time, it could lead to metering discrepancies for hundreds of users, triggering complaints and a crisis of trust.

[0008] More seriously, there is currently a lack of an effective evaluation system for the fusion effect. Traditional evaluation indicators such as sampling accuracy and communication packet loss rate only reflect the underlying performance and cannot answer the core question of "how fusion quality affects DSR effectiveness". For example, there is no model that can quantitatively explain: "If the data latency is reduced from 50ms to 10ms, how much will the overall response efficiency be improved?" or "If the device parameter calibration frequency is changed from once a day to once an hour, how many settlement disputes will be reduced?" - This makes system optimization lack a clear direction.

[0009] Although existing research has attempted to introduce wavelet transform and deep learning for multi-scale feature extraction, most of these studies remain at the laboratory stage, are not deeply coupled with DSR business logic, and lack value mapping of the "data fusion → control effect → economic / social benefits" chain.

[0010] Therefore, there is an urgent need for a new data fusion method that is oriented towards DSR business needs, takes into account multiple time scales, and supports benefit inversion, so as to break through the bottleneck of the current acquisition device that is "visible but not usable". Summary of the Invention

[0011] According to the present invention, a multi-source heterogeneous data fusion method and system oriented towards demand-side response are provided to overcome the three major problems of time scale mismatch, rigid fusion strategy and evaluation detachment from business when existing acquisition devices process multi-source heterogeneous data, so as to achieve a high degree of synergy between the data fusion process and DSR application scenarios, and improve response efficiency and user trust level.

[0012] According to a first aspect of the present invention, a multi-source heterogeneous data fusion method oriented towards demand-side response is provided, comprising:

[0013] Multi-source heterogeneous data is acquired from the plant's data acquisition device, including high-frequency power data, device parameter data, and DSR business indicator data.

[0014] Access the peak-valley-flat electricity price signal and DR event activation notification from the power grid dispatch center, identify the current DSR scenario type, and set differentiated fusion targets;

[0015] A dual-channel alignment model of "uplink aggregation + downlink prediction" is constructed. Based on the fusion target, the high-frequency power data is dimensionality-reduced and aligned to generate aligned high-frequency power data. The DSR service indicator data is dimensionality-up aligned to generate aligned DSR service indicator data.

[0016] Based on the current DSR scenario type, the aligned high-frequency power data, device parameter data, and aligned DSR business indicator data are merged to obtain a fused comprehensive dataset.

[0017] Optionally, the high-frequency power data includes: instantaneous values ​​of voltage u and current i, with a sampling interval of a predetermined millisecond;

[0018] Device parameter data includes: acquisition interval of a predetermined number of minutes, including CT / PT error and sampling clock offset;

[0019] DSR business metrics data include: data collection intervals of predetermined minutes or hours, including response volume and user engagement rate.

[0020] Optionally, identify the current DSR scenario type and set differentiated fusion targets; including:

[0021] The triggering conditions for identifying a peak-hour rapid response scenario are: being in a preset peak-hour period or the line load rate exceeding a predetermined value; the fusion goal is to control data latency within a predetermined range.

[0022] The triggering condition for identifying a fair settlement scenario during off-peak hours is: being in a preset off-peak time period and having no emergency DR events; its fusion goal is to control parameter errors within a predetermined range.

[0023] Optionally, a dual-channel alignment model of "uplink aggregation + downlink prediction" is constructed. Based on the fusion objective, the high-frequency power data is dimensionality-reduced and aligned to generate aligned high-frequency power data, and the DSR service indicator data is dimensionality-up aligned to generate aligned DSR service indicator data, including:

[0024] The high-frequency electrical energy data is dimensionality-reduced and aligned with the device parameter data to generate aligned high-frequency electrical energy data, specifically as follows:

[0025] For instantaneous voltage / current values ​​at the 0.02-second level, a sliding window RMS algorithm is used to aggregate them into minute-level feature quantities, which are then aligned with the device parameter data.

[0026]

[0027] Where N = 3000 is the number of sampling points within 1 minute, i(t) k () represents the instantaneous current value at the k-th sampling time;

[0028] For the DSR service indicator data, an LSTM time series prediction model is used to generate DSR service indicator data that is aligned with the high-frequency power data.

[0029] Optionally, based on the current DSR scenario type, the aligned high-frequency power data, device parameter data, and aligned DSR business indicator data are fused to obtain a fused comprehensive dataset, including:

[0030] Based on the fusion function, the aligned high-frequency power data, device parameter data, and aligned DSR business indicator data are fused to obtain the fused comprehensive dataset:

[0031] The fusion function is:

[0032] F(t)=α(t)·H(t)+β(t)·P(t)+γ(t)·B(t)

[0033] Where α+β+γ=1, the weight coefficients are dynamically generated by the scene controller according to the following rules:

[0034] Peak period: α>β≈γ, emphasizing real-time performance;

[0035] During the trough period: β increases significantly, with a focus on the risk of parameter drift;

[0036] H(t) represents the aligned high-frequency data, P(t) represents the device parameter data, and B(t) represents the aligned DSR index data.

[0037] According to another aspect of the present invention, a multi-source heterogeneous data fusion system oriented towards demand-side response is also provided, comprising:

[0038] A multi-source heterogeneous data acquisition module is used to acquire multi-source heterogeneous data from the power plant acquisition device. The data includes high-frequency power data, device parameter data, and DSR business indicator data.

[0039] The module for setting fusion targets is used to access peak-valley-flat electricity price signals and DR event activation notifications from the power grid dispatch center, identify the current DSR scenario type, and set differentiated fusion targets.

[0040] The multi-source heterogeneous data alignment module is used to construct a dual-channel alignment model of "uplink aggregation + downlink prediction". Based on the fusion target, the high-frequency power data is dimensionality-reduced and aligned to generate aligned high-frequency power data, and the DSR service indicator data is dimensionality-up aligned to generate aligned DSR service indicator data.

[0041] The integrated data module is used to merge aligned high-frequency power data, device parameter data, and aligned DSR business indicator data according to the current DSR scenario type to obtain a merged integrated dataset.

[0042] Optionally, the target module for fusion is defined, including:

[0043] The peak period identification submodule is used to identify the trigger conditions for peak period fast response scenarios as: being in a preset peak period interval or the line load rate exceeding a predetermined value. Its integration goal is to control the data latency within a predetermined range.

[0044] The triggering condition for identifying a fair settlement scenario during off-peak hours is: being in a preset off-peak time period and having no emergency DR events. The fusion goal is to control the parameter error within a predetermined range.

[0045] Optionally, aligning multi-source heterogeneous data modules includes:

[0046] The high-frequency power data alignment submodule is used to perform dimensionality reduction and alignment on the high-frequency power data, aligning it with the device parameter data to generate aligned high-frequency power data, specifically as follows:

[0047] For instantaneous voltage / current values ​​at the 0.02-second level, a sliding window RMS algorithm is used to aggregate them into minute-level feature quantities, which are then aligned with the device parameter data.

[0048]

[0049] Where N = 3000 is the number of sampling points within 1 minute, i(t) k () represents the instantaneous current value at the k-th sampling time;

[0050] The DSR service indicator data alignment submodule is used to generate DSR service indicator data aligned with the high-frequency power data using an LSTM time series prediction model.

[0051] Optionally, a fusion and integration data module is obtained, including:

[0052] The integrated data acquisition submodule is used to fuse aligned high-frequency power data, device parameter data, and aligned DSR business indicator data according to a fusion function to obtain a fused integrated dataset.

[0053] The fusion function is:

[0054] F(t)=α(t)·H(t)+β(t)·P(t)+γ(t)·B(t)

[0055] Where α+β+γ=1, the weight coefficients are dynamically generated by the scene controller according to the following rules:

[0056] Peak period: α>β≈γ, emphasizing real-time performance;

[0057] During the trough period: β increases significantly, with a focus on the risk of parameter drift;

[0058] H(t) represents the aligned high-frequency data, P(t) represents the device parameter data, and B(t) represents the aligned DSR index data.

[0059] According to another aspect of the invention, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method as described in any one of the claims.

[0060] According to another aspect of the present invention, an electronic device is also provided, comprising:

[0061] The aforementioned computer-readable storage medium; and

[0062] One or more processors for executing a program in the computer-readable storage medium.

[0063] Therefore, a data fusion method and system for demand-side response (DSR) is disclosed, aiming to solve the problems of time scale mismatch, difficulty in coordinating heterogeneous data types, and difficulty in quantifying fusion effects in the fusion of multi-source heterogeneous data from existing centralized data acquisition devices in power plants. The method is based on a three-layer architecture of "scenario-driven—multi-scale alignment—benefit mapping". First, data fusion tasks are divided according to DSR operation scenarios (second-level response during peak hours, fair settlement during off-peak hours). Second, a time scale conversion model is constructed to achieve spatiotemporal alignment of 0.02-second high-frequency energy data, minute-level device parameters, and 15-minute / hour-level DSR indicators. Third, a dynamic weighted fusion algorithm is introduced to coordinate the contribution of analog, digital, and statistical quantities in conjunction with scenario priorities. Finally, a fusion effect-DSR benefit correlation evaluation model is established to quantify the impact of data delay or error on response rate and user satisfaction. This invention improves the timeliness and business adaptability of data fusion, enhances the accuracy of DSR control and user trust, and is suitable for intelligent acquisition terminals in large-scale demand-side management systems. Attached Figure Description

[0064] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures:

[0065] Figure 1 This is a flowchart illustrating a multi-source heterogeneous data fusion method for demand-side response as described in this embodiment.

[0066] Figure 2 This is a schematic diagram of data fusion as described in this embodiment;

[0067] Figure 3 This is a schematic diagram of the structure of the data fusion system described in this embodiment;

[0068] Figure 4 This is a schematic diagram of a multi-source heterogeneous data fusion system oriented towards demand-side response as described in this embodiment. Detailed Implementation

[0069] Exemplary embodiments of the invention will now be described with reference to the accompanying drawings. However, the invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to fully and completely disclose the invention and to fully convey its scope to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the drawings is not intended to limit the invention. In the drawings, the same units / elements are referred to by the same reference numerals.

[0070] Unless otherwise stated, the terms used herein (including technical terms) have their common meaning as understood by one of ordinary skill in the art. Furthermore, it is understood that terms defined in commonly used dictionaries should be understood to have a meaning consistent with the context of their relevant field, and not to be interpreted as having an idealized or overly formal meaning.

[0071] According to a first aspect of the present invention, a multi-source heterogeneous data fusion method 100 oriented towards demand-side response is provided, with reference to Figure 1 As shown, the method 100 includes:

[0072] S101: Acquire multi-source heterogeneous data from the plant acquisition device, including high-frequency power data, device parameter data, and DSR business indicator data;

[0073] S102: Access the peak-valley-flat electricity price signal and DR event activation notification from the power grid dispatch center, identify the current DSR scenario type, and set differentiated fusion targets;

[0074] S103: Construct a dual-channel alignment model of "uplink aggregation + downlink prediction". Based on the fusion target, perform dimensionality reduction alignment on the high-frequency power data to generate aligned high-frequency power data, and perform dimensionality increase alignment on the DSR service indicator data to generate aligned DSR service indicator data.

[0075] S104: Based on the current DSR scenario type, merge the aligned high-frequency power data, device parameter data, and aligned DSR business indicator data to obtain a merged comprehensive dataset.

[0076] Specifically, this invention proposes a three-in-one data fusion architecture of "scenario-driven, multi-scale alignment, and benefit mapping" to efficiently fuse multi-source heterogeneous data (high-frequency power data, device parameters, and demand-side response (DSR) business indicators) collected by centralized data acquisition devices in power plants. The specific steps include the following steps.

[0077] Based on the "spatiotemporal dual-dimensional" identification of the current DSR scenario, the fusion task is divided into peak-hour rapid response type and valley-hour settlement fairness type, and a differentiated objective function is preset.

[0078] By accessing the power grid dispatch center's "peak-valley flat electricity price signal" and "DR event activation notification" (such as triggering an early warning when the 10kV line load rate exceeds 90%), the system automatically determines whether the current situation is:

[0079] Peak hours (e.g., 18:00-22:00): The objective function is to minimize data latency (≤10ms) to ensure the real-time performance of load shedding commands;

[0080] During off-peak hours (e.g., 00:00-08:00): the objective function is to maximize data reliability (parameter error ≤ 0.1%) and ensure settlement fairness.

[0081] This invention breaks through the limitations of the traditional "single sampling rate conversion" and constructs a dual-channel alignment model of "uplink aggregation + downlink prediction" to achieve seamless collaboration of data at the 20ms-hour level.

[0082] Uplink Channel: High-Frequency Data Dimensionality Reduction and Alignment

[0083] For instantaneous voltage / current values ​​in the 0.02-second range, a sliding window RMS algorithm is used to aggregate them into minute-level features, which are then aligned with device parameters (such as CT ratio and temperature drift coefficient).

[0084]

[0085] (where N = 3000 is the number of sampling points within 1 minute, i(t) k (where ) is the instantaneous current value at the k-th sampling time.

[0086] Downward trend: Upgrading and supplementing low-frequency indicators

[0087] For 15-minute / hourly DSR metrics (such as response capacity and user engagement rate), an LSTM time series prediction model is used to generate high-frequency context information (such as outputting a prediction value every 100ms) to solve the problem of "no latest metric reference when decision instructions are issued".

[0088] The design incorporates a scenario-adaptive weighting mechanism that adjusts the contribution of high-frequency data (H), device parameters (P), and DSR indicators (B) in real time to achieve "precise adaptation to peak and valley scenarios".

[0089] Fusion function:

[0090] F(t)=α(t)·H(t)+β(t)·P(t)+γ(t)·B(t)

[0091] Where α+β+γ=1, the weight coefficients are dynamically generated by the scene controller according to the following rules:

[0092] Peak time period: α>β≈γ, emphasizing real-time performance.

[0093] During the trough period: β increases significantly, so pay close attention to the risk of parameter drift.

[0094] When a “DR event level upgrade” is detected (e.g., from “voluntary response” to “mandatory reduction”), α automatically increases by 0.1; when the device parameter verification result shows “error exceeds threshold” (e.g., CT ratio error > 0.3%), β is temporarily increased to 0.8 to prioritize data reliability.

[0095] Compared with the prior art, the present invention has the following advantages:

[0096] (1) Improve response timeliness: Through scenario-based fusion strategy optimization, the effective data latency can be controlled within 10ms during peak periods, and the response command issuance speed can be increased by more than 40%.

[0097] (2) Enhance the credibility of settlement: Actively identify abnormal device parameters during off-peak hours and provide early warning of potential metering deviations, which is expected to reduce the user complaint rate by 60%;

[0098] (3) Achieve quantification of integrated value: For the first time, a causal chain model of "data quality → DSR performance" is established to provide data basis for system upgrades;

[0099] (4) Strong compatibility: It can be implemented through software upgrades based on existing data acquisition terminals without the need to replace hardware;

[0100] (5) High degree of automation: The fusion strategy adapts to switch, reducing the burden of manual intervention.

[0101] The apparatus of the present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly describe the technical solution of the present invention and should not be construed as limiting the scope of protection of this application.

[0102] Figure 2 The diagram illustrates data fusion. The system monitors all demand-side data to determine if it has entered a peak / valley DSR period. It then acquires high-frequency power data (20ms intervals, including instantaneous u / i values), device parameter data (1-minute intervals, including CT / PT errors and sampling clock offset), and DSR business indicators (15 minutes / hour level, including response volume and participation rate) from the power plant's data acquisition devices. Outlier removal, unit normalization, and timestamp calibration are performed on the raw data. Finally, the fused comprehensive dataset is pushed to the EMS system for real-time control.

[0103] Figure 3The diagram shows the structural block of the data fusion system. This flowchart illustrates the technical architecture under the DSP business scenario, covering the application layer, fusion layer, data layer, and benefit evaluation layer. The application layer sets two business objectives: rapid response during peak hours (target latency ≤ 10ms) and fair settlement during off-peak hours (target error ≤ 0.1%). The fusion layer adjusts weights in real time according to the scenario through a dynamic weight controller. High-frequency data (H) has an α of 0.7 during peak hours and α of 0.2 during off-peak hours; device parameters (P) have an β of 0.15 during peak hours and β of 0.6 during off-peak hours; and DSP indicators (B) have an γ of 0.15 during peak hours and γ of 0.2 during off-peak hours. The data layer is divided into uplink and downlink channels. The uplink channel (high-frequency dimensionality reduction) uses a sliding window effective value algorithm. The 20ms instantaneous value is transformed into minute-level feature quantity. The downlink channel (low-frequency upscaling) uses the LSTM time series prediction model to transform the 15-minute / hour-level indicators into 100ms-level predicted values. The benefit evaluation layer uses data latency (ms), parameter error (%), and packet loss rate (%) as inputs and response compliance rate (%), settlement dispute rate (%), and user satisfaction as outputs to evaluate the system's benefits. It fully presents the entire process from establishing business objectives to weight fusion, data processing, and benefit evaluation, reflecting the refined management and optimization of DSP business scenarios.

[0104] The hardware deployment of this invention adopts an "edge-cloud collaborative" distributed architecture, which is adapted to the characteristics of "massive terminals + centralized decision-making" in demand-side response (DSR) scenarios, specifically including:

[0105] The system utilizes an ARM Cortex-A53 quad-core processor (1.2GHz), which supports hardware-level real-time task scheduling. It integrates 8 analog inputs (AI) and 4 digital inputs (DI), enabling high-frequency sampling of voltage / current signals (0-10V / 4-20mA) at a sampling rate of 20ms / sample (50Hz power frequency scenario) with an accuracy of 0.2S. It also acquires status signals such as switch positions (remote signaling) and device operating light status, with a response time ≤1ms. Finally, it employs dual Ethernet ports (100Mbps) + a 4G module (supporting CAT-1 protocol) to achieve data transmission with a "wired primary, wireless backup" configuration.

[0106] Furthermore, high-frequency data from up to 100 terminals is aggregated in real time to reduce cloud transmission pressure (data compression ratio reaches 1:20). A three-level DSR event response strategy is pre-stored, allowing for independent execution of a basic response based on local data when cloud communication is interrupted (sustainability ≥ 2 hours). Finally, an industrial-grade embedded gateway is used, supporting a wide temperature range of -40℃ to 70℃, with a mean time between failures (MTBF) ≥ 50,000 hours.

[0107] like Figure 3As shown in the data layer, the core algorithm of this invention is realized through a five-step closed loop of "data preprocessing → scene recognition → multi-scale alignment → dynamic fusion → benefit feedback". The current signal (i(t)) sampled for 20ms is filtered by moving average to remove glitches and interference (window size N=5). The switch state (0 / 1) and alarm signal (text) are converted into numerical codes, such as "switch closed" → 1, "switch open" → 0; "device normal" → 0, "sampling abnormal" → 1, "communication interruption" → 2.

[0108] Multi-scale data alignment is performed at the cloud layer, comprising uplink aggregation (high frequency → low frequency) and downlink prediction (low frequency → high frequency). Uplink aggregation is performed on preprocessed 20ms current data (i... raw The sliding window effective value algorithm is used to calculate minute-level features (RMS). min The calculation formula is: Where 3000 represents the number of sampling points within one minute (60 seconds / 0.02 seconds / sample). Downlink prediction targets 15-minute DSR indicators (such as "current available load capacity to be reduced"), using an LSTM neural network to generate 100ms-level prediction values. The network structure is as follows: the input layer is the indicator value (x) from the previous 3 cycles. t-15 x t-30 x t-45 The hidden layer consists of two LSTM layers (64 neurons each) plus Dropout (to prevent overfitting); the output layer contains 600 predicted points for the next minute (one value every 100ms).

[0109] Furthermore, dynamic weight fusion is performed at the cloud layer. The weight coefficient generation rule is to dynamically adjust the weights (α, β, γ) of high-frequency data (H), device parameters (P), and DSR indicators (B) based on the scenario label. Specifically, under the peak-emergency response scenario label, α (H weight) is 0.7, β (P weight) is 0.15, and γ (B weight) is 0.15, triggered when the event level is 3 or the line load rate is >90%; under the off-peak-settlement verification scenario label, α is 0.2, β is 0.6, and γ is 0.2, triggered when the time period is off-peak and there is no emergency event; under the normal-regular monitoring scenario label, α is 0.5, β is 0.3, and γ is 0.2, triggered when it is a non-peak / off-peak time period; under the anomaly compensation mode scenario label, α is 0.1, β is 0.8, and γ is 0.1, triggered when the device parameter error is >0.3% (triggering the calibration mechanism). Taking the peak-emergency response scenario as an example, the fusion result F(t) = 0.7·H(t) + 0.15·P(t) + 0.15·B(t), where H(t) is the current sampling value of the current in the current 20ms (after preprocessing), P(t) is the CT ratio of the most recently calibrated CT (e.g., 1000 / 5A), and B(t) is the target response capacity of the current DR event (e.g., 500kW).

[0110] Benefit mapping and feedback optimization mainly includes two parts: data collection and optimization strategy generation. Regarding data collection, technical indicators for each fusion process are recorded through data tracking, such as the time difference from sampling to fusion completion (data latency) and parameter errors compared to the standard source; as well as business indicators, such as the actual user response speed (in seconds) after the DSR command is issued and the user confirmation rate of the settlement order (expressed as a percentage). Regarding optimization strategy generation, specific optimization suggestions are output based on the benefit mapping model (such as multiple linear regression).

[0111] This solves three core problems existing in centralized data acquisition devices at plants and stations when supporting DSR applications: poor data timeliness, disconnect between fusion strategies and business scenarios, and gap between technical indicators and business value.

[0112] Optionally, the high-frequency power data includes: instantaneous values ​​of voltage u and current i, with a sampling interval of a predetermined millisecond;

[0113] Device parameter data includes: acquisition interval of a predetermined number of minutes, including CT / PT error and sampling clock offset;

[0114] DSR business metrics data include: data collection intervals of predetermined minutes or hours, including response volume and user engagement rate.

[0115] Optionally, identify the current DSR scenario type and set differentiated fusion targets; including:

[0116] The triggering conditions for identifying a peak-hour rapid response scenario are: being in a preset peak-hour period or the line load rate exceeding a predetermined value; the fusion goal is to control data latency within a predetermined range.

[0117] The triggering condition for identifying a fair settlement scenario during off-peak hours is: being in a preset off-peak time period and having no emergency DR events; its fusion goal is to control parameter errors within a predetermined range.

[0118] Optionally, a dual-channel alignment model of "uplink aggregation + downlink prediction" is constructed. Based on the fusion objective, the high-frequency power data is dimensionality-reduced and aligned to generate aligned high-frequency power data, and the DSR service indicator data is dimensionality-up aligned to generate aligned DSR service indicator data, including:

[0119] The high-frequency electrical energy data is dimensionality-reduced and aligned with the device parameter data to generate aligned high-frequency electrical energy data, specifically as follows:

[0120] For instantaneous voltage / current values ​​at the 0.02-second level, a sliding window RMS algorithm is used to aggregate them into minute-level feature quantities, which are then aligned with the device parameter data.

[0121]

[0122] Where N = 3000 is the number of sampling points within 1 minute, i(t) k () represents the instantaneous current value at the k-th sampling time;

[0123] For the DSR service indicator data, an LSTM time series prediction model is used to generate DSR service indicator data that is aligned with the high-frequency power data.

[0124] Optionally, based on the current DSR scenario type, the aligned high-frequency power data, device parameter data, and aligned DSR business indicator data are fused to obtain a fused comprehensive dataset, including:

[0125] Based on the fusion function, the aligned high-frequency power data, device parameter data, and aligned DSR business indicator data are fused to obtain the fused comprehensive dataset:

[0126] The fusion function is:

[0127] F(t)=α(t)·H(t)+β(t)·P(t)+γ(t)·B(t)

[0128] Where α+β+γ=1, the weight coefficients are dynamically generated by the scene controller according to the following rules:

[0129] Peak period: α>β≈γ, emphasizing real-time performance;

[0130] During the trough period: β increases significantly, with a focus on the risk of parameter drift;

[0131] H(t) represents the aligned high-frequency data, P(t) represents the device parameter data, and B(t) represents the aligned DSR index data.

[0132] This invention addresses the problems of time-scale mismatch, difficulty in coordinating heterogeneous data types, and difficulty in quantifying fusion effects in existing centralized data acquisition devices for power plants during multi-source heterogeneous data fusion. The method is based on a three-layer architecture of "scenario-driven—multi-scale alignment—benefit mapping." First, it divides data fusion tasks according to DSR operating scenarios (second-level response during peak hours, fair settlement during off-peak hours). Second, it constructs a time-scale conversion model to achieve spatiotemporal alignment of 0.02-second high-frequency energy data, minute-level device parameters, and 15-minute / hour-level DSR indicators. Third, it introduces a dynamic weighted fusion algorithm to coordinate the contributions of analog, digital, and statistical quantities based on scenario priorities. Finally, it establishes a fusion effect-DSR benefit correlation evaluation model to quantify the impact of data latency or error on response rate and user satisfaction. This invention improves the timeliness and business adaptability of data fusion, enhances DSR control accuracy and user trust, and is suitable for intelligent acquisition terminals in large-scale demand-side management systems.

[0133] According to another aspect of the present invention, a multi-source heterogeneous data fusion system 400 oriented towards demand-side response is also provided, the system 400 comprising:

[0134] The multi-source heterogeneous data acquisition module 410 is used to acquire multi-source heterogeneous data from the power plant acquisition device. The data includes high-frequency power data, device parameter data, and DSR business indicator data.

[0135] The fusion target module 420 is used to access the peak-valley-flat electricity price signal and DR event activation notification from the power grid dispatch center, identify the current DSR scenario type, and set differentiated fusion targets.

[0136] The multi-source heterogeneous data alignment module 430 is used to construct a dual-channel alignment model of "uplink aggregation + downlink prediction". Based on the fusion target, the high-frequency power data is dimensionality-reduced and aligned to generate aligned high-frequency power data, and the DSR service indicator data is dimensionality-up aligned to generate aligned DSR service indicator data.

[0137] The integrated data module 440 is used to merge aligned high-frequency power data, device parameter data, and aligned DSR business indicator data according to the current DSR scenario type to obtain an integrated dataset.

[0138] Optionally, the target module for fusion is defined, including:

[0139] The peak period identification submodule is used to identify the trigger conditions for peak period fast response scenarios as: being in a preset peak period interval or the line load rate exceeding a predetermined value. Its integration goal is to control the data latency within a predetermined range.

[0140] The triggering condition for identifying a fair settlement scenario during off-peak hours is: being in a preset off-peak time period and having no emergency DR events. The fusion goal is to control the parameter error within a predetermined range.

[0141] Optionally, aligning multi-source heterogeneous data modules includes:

[0142] The high-frequency power data alignment submodule is used to perform dimensionality reduction and alignment on the high-frequency power data, aligning it with the device parameter data to generate aligned high-frequency power data, specifically as follows:

[0143] For instantaneous voltage / current values ​​at the 0.02-second level, a sliding window RMS algorithm is used to aggregate them into minute-level feature quantities, which are then aligned with the device parameter data.

[0144]

[0145] Where N = 3000 is the number of sampling points within 1 minute, i(t) k () represents the instantaneous current value at the k-th sampling time;

[0146] The DSR service indicator data alignment submodule is used to generate DSR service indicator data aligned with the high-frequency power data using an LSTM time series prediction model.

[0147] Optionally, a fusion and integration data module is obtained, including:

[0148] The integrated data acquisition submodule is used to fuse aligned high-frequency power data, device parameter data, and aligned DSR business indicator data according to a fusion function to obtain a fused integrated dataset.

[0149] The fusion function is:

[0150] F(t)=α(t)·H(t)+β(t)·P(t)+γ(t)·B(t)

[0151] Where α+β+γ=1, the weight coefficients are dynamically generated by the scene controller according to the following rules:

[0152] Peak period: α>β≈γ, emphasizing real-time performance;

[0153] During the trough period: β increases significantly, with a focus on the risk of parameter drift;

[0154] H(t) represents the aligned high-frequency data, P(t) represents the device parameter data, and B(t) represents the aligned DSR index data.

[0155] The multi-source heterogeneous data fusion system 400 oriented to demand-side response according to an embodiment of the present invention corresponds to the multi-source heterogeneous data fusion method 100 oriented to demand-side response according to another embodiment of the present invention, and will not be described again here.

[0156] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0157] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0158] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0159] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0160] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0161] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A multi-source heterogeneous data fusion method oriented towards demand-side response, characterized in that, include: Multi-source heterogeneous data is acquired from the plant's data acquisition device, including high-frequency power data, device parameter data, and DSR business indicator data. Access the peak-valley-flat electricity price signal and DR event activation notification from the power grid dispatch center, identify the current DSR scenario type, and set differentiated fusion targets; A dual-channel alignment model of "uplink aggregation + downlink prediction" is constructed. Based on the fusion target, the high-frequency power data is dimensionality-reduced and aligned to generate aligned high-frequency power data. The DSR service indicator data is dimensionality-up aligned to generate aligned DSR service indicator data. Based on the current DSR scenario type, the aligned high-frequency power data, device parameter data, and aligned DSR business indicator data are merged to obtain a fused comprehensive dataset.

2. The method according to claim 1, characterized in that, The high-frequency power data includes: instantaneous values ​​of voltage u and current i, with a sampling interval of a predetermined millisecond. Device parameter data includes: acquisition interval of a predetermined number of minutes, including CT / PT error and sampling clock offset; DSR business metrics data include: data collected at predetermined intervals of minutes or hours, including response volume and user engagement rate.

3. The method according to claim 1, characterized in that, Identify the current DSR scenario type and set differentiated fusion objectives; including: The triggering conditions for identifying a peak-hour rapid response scenario are: being in a preset peak-hour period or the line load rate exceeding a predetermined value; the fusion goal is to control data latency within a predetermined range. The triggering condition for identifying a fair settlement scenario during off-peak hours is: being in a preset off-peak time period and having no emergency DR events; its fusion goal is to control parameter errors within a predetermined range.

4. The method according to claim 1, characterized in that, A dual-channel alignment model of "uplink aggregation + downlink prediction" is constructed. Based on the fusion objective, the high-frequency power data is dimensionality-reduced and aligned to generate aligned high-frequency power data. The DSR service indicator data is dimensionality-up aligned to generate aligned DSR service indicator data, including: The high-frequency electrical energy data is dimensionality-reduced and aligned with the device parameter data to generate aligned high-frequency electrical energy data, specifically as follows: For instantaneous voltage / current values ​​at the 0.02-second level, a sliding window RMS algorithm is used to aggregate them into minute-level feature quantities, which are then aligned with the device parameter data. Where N = 3000 is the number of sampling points within 1 minute, i(t) k () represents the instantaneous current value at the k-th sampling time; For the DSR service indicator data, an LSTM time series prediction model is used to generate DSR service indicator data aligned with the high-frequency power data.

5. The method according to claim 1, characterized in that, Based on the current DSR scenario type, the aligned high-frequency power data, device parameter data, and aligned DSR business indicator data are fused to obtain a fused comprehensive dataset, including: Based on the fusion function, the aligned high-frequency power data, device parameter data, and aligned DSR service indicator data are fused to obtain the fused comprehensive dataset: The fusion function is: F(t)=α(t)·H(t)+β(t)·P(t)+γ(t)·B(t) Where α+β+γ=1, the weight coefficients are dynamically generated by the scene controller according to the following rules: Peak period: α>β≈γ, emphasizing real-time performance; During the trough period: β increases significantly, with a focus on the risk of parameter drift; H(t) represents the aligned high-frequency data, P(t) represents the device parameter data, and B(t) represents the aligned DSR index data.

6. A multi-source heterogeneous data fusion system oriented towards demand-side response, characterized in that, include: A multi-source heterogeneous data acquisition module is used to acquire multi-source heterogeneous data from the power plant acquisition device. The data includes high-frequency power data, device parameter data, and DSR business indicator data. The module for setting fusion targets is used to access peak-valley-flat electricity price signals and DR event activation notifications from the power grid dispatch center, identify the current DSR scenario type, and set differentiated fusion targets. The multi-source heterogeneous data alignment module is used to construct a dual-channel alignment model of "uplink aggregation + downlink prediction". Based on the fusion target, the high-frequency power data is dimensionality-reduced and aligned to generate aligned high-frequency power data, and the DSR service indicator data is dimensionality-up aligned to generate aligned DSR service indicator data. The integrated data module is used to merge aligned high-frequency power data, device parameter data, and aligned DSR business indicator data according to the current DSR scenario type to obtain a merged integrated dataset.

7. The system according to claim 6, characterized in that, Define the target fusion module, including: The peak period identification submodule is used to identify the trigger conditions for peak period fast response scenarios as: being in a preset peak period interval or the line load rate exceeding a predetermined value. Its integration goal is to control the data latency within a predetermined range. The triggering condition for identifying a fair settlement scenario during off-peak hours is: being in a preset off-peak time period and having no emergency DR events. The fusion goal is to control the parameter error within a predetermined range.

8. The system according to claim 7, characterized in that, Alignment of multi-source heterogeneous data modules, including: The high-frequency power data alignment submodule is used to perform dimensionality reduction and alignment on the high-frequency power data, aligning it with the device parameter data to generate aligned high-frequency power data, specifically as follows: For instantaneous voltage / current values ​​at the 0.02-second level, a sliding window RMS algorithm is used to aggregate them into minute-level feature quantities, which are then aligned with the device parameter data. Where N = 3000 is the number of sampling points within 1 minute, i(t) k () represents the instantaneous current value at the k-th sampling time; The DSR service indicator data alignment submodule is used to generate DSR service indicator data aligned with the high-frequency power data using an LSTM time series prediction model.

9. The system according to claim 6, characterized in that, Obtain the integrated data module, including: The integrated data acquisition submodule is used to fuse aligned high-frequency power data, device parameter data, and aligned DSR business indicator data according to a fusion function to obtain a fused integrated dataset. The fusion function is: F(t)=α(t)·H(t)+β(t)·P(t)+γ(t)·B(t) Where α+β+γ=1, the weight coefficients are dynamically generated by the scene controller according to the following rules: Peak period: α>β≈γ, emphasizing real-time performance; During the trough period: β increases significantly, with a focus on the risk of parameter drift; H(t) represents the aligned high-frequency data, P(t) represents the device parameter data, and B(t) represents the aligned DSR index data.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-5.

11. An electronic device, characterized in that, include: The computer-readable storage medium as described in claim 10; as well as One or more processors for executing a program in the computer-readable storage medium.

Images