An intelligent operation and maintenance management system for new energy equipment based on an internet of things

Abstract

The application discloses a new energy equipment intelligent operation and maintenance management system based on Internet of Things, and particularly relates to the field of equipment intelligence operation and maintenance, and aims to solve the problem that the equipment operation control strategy deviation is difficult to automatically repair in the prior art. A control instruction in an operation process is collected through a track extraction module to construct a two-dimensional behavior track matrix; a disturbance mapping module is used to establish a corresponding relationship between a disturbance factor and a behavior track; an offset identification module is used to determine whether a control path deviates from an original operation mode after the disturbance ends; when a track deviation is detected, an inertia identification module is called to extract an evolution trend of a tail track; an offset identification model is constructed by a model training module to determine whether a control system has a benchmark memory deviation; finally, a parameter rollback operation is performed through a reset maintenance module to automatically restore the control state, which can significantly improve the operation and maintenance efficiency and system stability of the new energy equipment.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for equipment, and more specifically, to an intelligent operation and maintenance management system for new energy equipment based on the Internet of Things. Background Technology

[0002] New energy equipment such as wind power, photovoltaics, and energy storage are widely equipped with environmental adaptive control systems under the Internet of Things (IoT) architecture to respond in real time to external disturbances, such as sudden wind speed changes, shading, and grid voltage fluctuations. Common control response mechanisms include pitch regulation for wind turbines, maximum power point tracking (MPPT) adjustment for photovoltaic inverters, and dynamic charge and discharge scheduling for energy storage systems. These mechanisms are based on the common assumption that environmental disturbances are short-term, transient, and quickly recoverable, thus the control system automatically returns to the equipment's design baseline state after the disturbance ends. However, in actual operation and maintenance scenarios, new energy equipment is frequently exposed to long-term extreme disturbance environments, such as sandstorms lasting for many hours, continuous shading caused by continuous rain, and charge and discharge limitations caused by continuous voltage disturbances. In such cases, the adaptive system will perform frequent compensation behaviors for a long time, and its short-term adjustment actions gradually solidify, evolving into a behavioral inertia deviation of the control system. Ultimately, this manifests as the equipment failing to return to its design conditions after the disturbance ends, instead maintaining operation in the compensated state, and mistakenly recording this state as the new stable operating baseline. Because the operation and maintenance platform relies on set parameter ranges to determine the operating status of equipment, it cannot identify this type of control parameter deviation degradation, resulting in a series of chain reactions such as hidden performance degradation, reduced policy adaptability, and abnormal decline in overall efficiency.

[0003] This type of problem is characterized by a high degree of behavioral layer structure, parameter layer deviation, and manifestation layer concealment. It is different from both physical damage and parameter anomalies. Therefore, traditional intelligent operation and maintenance strategies lack effective identification and intervention mechanisms. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an intelligent operation and maintenance management system for new energy equipment based on the Internet of Things to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] An intelligent operation and maintenance management system for new energy equipment based on the Internet of Things includes:

[0007] The trajectory extraction module is used to collect the control command sequence during the operation of new energy equipment and construct a two-dimensional behavior trajectory matrix that reflects the timing of control actions.

[0008] The perturbation mapping module is used to collect perturbation factor data and establish a mapping relationship between perturbation factors and behavioral trajectories.

[0009] The offset recognition module is used to identify whether the tail trajectory deviates from the initial trajectory based on the tail trajectory state after the disturbance phase ends, and to record the offset path of the corresponding control channel.

[0010] The inertial recognition module is used to extract the multi-channel evolution trend vector of the tail segment trajectory and construct a channel-level inertial trend structure when trajectory feature offset is detected.

[0011] The model training module is used to train the trajectory trend recognition model. It combines the inertial trend structure and the control strategy output to form the feature input and determine whether the control strategy has a reference memory offset.

[0012] The reset maintenance module is used to construct an automatic reset command sequence based on the identification result when the reference offset state is identified, and to send parameter rollback instructions to the control strategy.

[0013] In a preferred embodiment, the trajectory extraction module collects the control command sequence during the operation of the new energy equipment and constructs a two-dimensional behavioral trajectory matrix reflecting the timing of control actions, specifically including:

[0014] Collect all control commands of new energy equipment within the disturbance response cycle, and extract the control parameters, control channel identifier, feedback response delay and corresponding execution timestamp from each command;

[0015] The extracted instruction sequence is arranged in chronological order and divided into channel groups according to control channels;

[0016] Within each channel group, control action segments are divided based on a sliding window with a fixed time width. The magnitude of control parameter changes, the number of control triggers, and the average feedback response delay within each segment are statistically analyzed and integrated into a channel-level control behavior vector.

[0017] The behavior vectors of each channel are concatenated in chronological order to construct a structured time-channel two-dimensional behavior trajectory matrix.

[0018] In a preferred embodiment, the disturbance mapping module, specifically including the following steps: collecting disturbance factor data and establishing the mapping relationship between disturbance factors and behavioral trajectories:

[0019] Disturbance factor data associated with new energy equipment is extracted from the Internet of Things sensing network, and the disturbance factors correspond to the functional types of new energy equipment;

[0020] Based on the first derivative of the rate of change of the disturbance factor data and the disturbance amplitude threshold, the disturbance initiation point and the disturbance mitigation point are identified, and the disturbance is divided into three stages: initiation period, stable period and decay period according to the duration and change pattern of continuous disturbance.

[0021] The identified perturbation phase labels are mapped to the time axis of the behavior trajectory matrix according to the time alignment method;

[0022] Record the parameter changes, control channel trigger frequency, and local statistical characteristics of the behavior trajectory segment corresponding to each disturbance stage;

[0023] The statistical characteristics of the disturbance stage labels, the change rate of the disturbance factor data, and the behavioral trajectory segments are uniformly stored as a labeled disturbance response dataset.

[0024] In a preferred embodiment, the offset recognition module identifies whether the tail trajectory deviates from the initial trajectory based on the tail trajectory state after the disturbance phase ends, and records the offset path of the corresponding control channel, specifically including:

[0025] Extract a fixed-length segment of the behavior trajectory from the behavior trajectory matrix after the end of the disturbance decay period as the tail trajectory.

[0026] Extract the control parameters of the tail trajectory in each channel, and statistically analyze the activation order, control parameter variation range and feedback response delay consistency of each channel to construct the tail trajectory state vector;

[0027] Call the initial behavior trajectory segment with the same length as the tail trajectory before the disturbance import point, extract the initial behavior trajectory state vector using trajectory compression and alignment, and perform multi-dimensional channel vector difference calculation with the tail trajectory state vector.

[0028] According to the preset stability index range, the mean and variance stability tests are performed on the difference sequence to determine whether the tail trajectory state has deviated from the feature space of the initial behavior trajectory. If so, it is marked as trajectory feature offset.

[0029] When the tail segment trajectory has trajectory feature offset in at least a set number of channel dimensions, the corresponding tail segment trajectory is marked as an offset path, and the offset channel and offset direction are recorded.

[0030] In a preferred embodiment, the inertial recognition module, upon detecting a trajectory feature shift, extracts a multi-channel evolution trend vector of the tail segment trajectory and constructs a channel-level inertial trend structure, specifically including:

[0031] Extract the time evolution curves of the control parameters output for each channel from the offset path that has been marked as trajectory feature offset state;

[0032] In the control parameter change curves of each channel, an evolution trend sequence is constructed based on the change in curve slope, and the channel evolution trends under different perturbations are grouped according to the perturbation stage label.

[0033] Clustering methods are used to establish a path change pattern classification set for the trend vector sequence. Curvature similarity is performed on the control parameter curves corresponding to the channels within the group, and trend center templates are extracted to construct a trend classification space.

[0034] Multiple tail trajectories are mapped in the trend classification space to identify whether they converge to a specific trend center template. If the corresponding convergence path appears in multiple channels at the same time, the convergence path is marked as a behavioral inertia trend structure.

[0035] In a preferred embodiment, the model training module trains a trajectory trend recognition model, using the inertial trend structure and the control strategy output together as feature inputs. Determining whether the control strategy has a baseline memory offset specifically includes:

[0036] The channel and path convergence period and the slope of the control parameter change curve extracted from the behavioral inertia trend structure are used as feature inputs.

[0037] By combining the disturbance response dataset with the policy output parameters of the automatic control unit, training samples in a unified format are constructed.

[0038] A classification model construction method based on temporal residual aggregation mechanism is adopted, and the training samples are cross-trained in multiple rounds to establish a trajectory trend recognition model;

[0039] Input the tail trajectory features to be judged into the trajectory trend recognition model, calculate the degree of fit of the output results to each offset path, and output the confidence label of the potential benchmark memory offset.

[0040] In a preferred embodiment, the automatic control unit refers to the component in the new energy equipment that automatically executes decision control, and its corresponding strategy output parameters are control parameters automatically generated by the control program embedded in the automatic control unit or by the strategy logic.

[0041] In a preferred embodiment, the reset maintenance module, when identified as a reference offset state, constructs an automatic reset command sequence based on the identification result by tracing back the adjustment path, and issues a parameter rollback instruction to the control strategy, specifically including:

[0042] Obtain the potential baseline memory offset confidence label output by the trajectory trend recognition model, and perform confidence threshold determination on the corresponding channel;

[0043] In channels where the confidence level exceeds the preset offset threshold, the current control state of the channel is marked as parameter offset state;

[0044] The backtracking channel extracts the compensation path of the correction parameters by tracing all the adjustment command sequences from the disturbance introduction point to the current time.

[0045] Identify the maximum offset gradient point of the correction parameters in the compensation path, use the parameter of the previous stable segment before this point as the proposed reset reference value to construct an automatic reset adjustment command sequence, and issue a reset command to the controlled channel to execute the strategy output parameter rollback.

[0046] The technical effects and advantages of the intelligent operation and maintenance management system for new energy equipment based on the Internet of Things (IoT) of this invention are as follows:

[0047] This invention constructs a time-channel two-dimensional behavioral trajectory matrix through a trajectory extraction module, comprehensively reflecting the control behavior characteristics of new energy equipment within the disturbance response cycle. Combined with a disturbance mapping module, it establishes a temporal correlation between disturbance factors and behavioral trajectories, achieving precise labeling of the disturbance phase and positioning of the behavioral impact range. An offset identification module automatically identifies multi-channel state differences between the tail trajectory and the initial trajectory after disturbance attenuation, forming a structured judgment of control behavior feature drift. Furthermore, an inertial identification module extracts multi-channel evolution trends, establishing a trend classification space and inertial trend structure, improving the ability to identify fixed adjustment behaviors. A model training module integrates control strategy output and trend structure to construct multi-dimensional feature inputs, training a trajectory trend identification model to achieve intelligent discrimination of baseline memory offsets. Finally, a reset and maintenance module backtracks the adjustment path based on the trajectory trend identification model results, accurately identifying parameter offset positions and constructing an automatic reset command sequence, ensuring the targeting and stability of the control strategy parameter backtracking process. The overall solution overcomes the limitations of traditional static parameter comparison methods, achieving dynamic identification of disturbance behavior offsets and intelligent closed-loop parameter reset control. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the structure of an intelligent operation and maintenance management system for new energy equipment based on the Internet of Things according to the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0050] Example 1, Figure 1 This invention discloses an intelligent operation and maintenance management system for new energy equipment based on the Internet of Things, which includes:

[0051] The trajectory extraction module is used to collect the control command sequence during the operation of new energy equipment and construct a two-dimensional behavior trajectory matrix that reflects the timing of control actions.

[0052] The perturbation mapping module is used to collect perturbation factor data and establish a mapping relationship between perturbation factors and behavioral trajectories.

[0053] The offset recognition module is used to identify whether the tail trajectory deviates from the initial trajectory based on the tail trajectory state after the disturbance phase ends, and to record the offset path of the corresponding control channel.

[0054] The inertial recognition module is used to extract the multi-channel evolution trend vector of the tail segment trajectory and construct a channel-level inertial trend structure when trajectory feature offset is detected.

[0055] The model training module is used to train the trajectory trend recognition model. It combines the inertial trend structure and the control strategy output to form the feature input and determine whether the control strategy has a reference memory offset.

[0056] The reset maintenance module is used to construct an automatic reset command sequence based on the identification result when the reference offset state is identified, and to send parameter rollback instructions to the control strategy.

[0057] The trajectory extraction module collects the control command sequence during the operation of new energy equipment and constructs a two-dimensional behavioral trajectory matrix that reflects the timing of control actions.

[0058] After the new energy equipment enters the disturbance response cycle, all control commands issued by the equipment control layer are recorded, and the key structural fields of each command are identified. Control commands typically consist of an execution identifier, target channel number, control parameter value, control issuance time, and its corresponding feedback response time. In practice, the control data acquisition unit first interfaces with the command recording channel of the equipment controller to extract continuous raw control data and convert it into a structured command recording sequence.

[0059] Each instruction must strictly extract the following field information:

[0060] (1) Control parameters: These represent the set values ​​used for adjustment at that moment, such as output power set value, voltage control value, inverter frequency set value, etc.

[0061] (2) Control channel identification: Clearly identify which actuator this parameter applies to, such as motor control unit, inverter power module, temperature control valve, etc., to ensure the accuracy of subsequent channel-level grouping;

[0062] (3) Feedback response delay: defined as the time interval between the issuance of the control command and the detection of the corresponding feedback signal change. The unit can be set to milliseconds. The response delay reflects the coupling efficiency between control, execution and feedback.

[0063] (4) Execution timestamp: Records the precise time point when the control command is issued, which is used for subsequent timing reconstruction and sliding window positioning.

[0064] Based on the structured sequence of control instructions with extracted fields, the data is further sorted uniformly along the timeline and grouped by control channel to form independent channel behavior data streams. In specific operations, all control instructions are first sorted in ascending order based on the execution timestamp field to reconstruct the complete instruction timing chain, ensuring the temporal continuity and logical consistency of control behavior.

[0065] After the time series reconstruction is completed, all control commands are assigned to their corresponding channel groups based on the control channel identifier field. Each channel group represents all command behaviors of the device on that control path, including characteristic sequences such as control value adjustment, trigger frequency, and response delay. For example, for a multi-channel control new energy inverter, its control channels may include output voltage regulation, frequency conversion control, temperature control fan control, etc. The system will classify the control commands into independent groups such as channel A, channel B, and channel C according to the channel number.

[0066] After channel grouping, a sliding window mechanism is introduced within each channel group to extract control pattern features with temporal locality and behavioral integrity from continuous control commands within each channel. The sliding window is set to a fixed time width, in milliseconds or seconds, and it is recommended to select a window width within 1 to 2 times the typical device response frequency. For example, for a device with a control frequency of 1Hz, a sliding window of 2 seconds and a window step of 1 second can be set to ensure timing continuity and detail coverage.

[0067] Within each sliding window, the system extracts all control commands within that time period and sequentially calculates the following behavioral metrics:

[0068] (1) Control parameter variation range: Calculate the difference between the maximum and minimum values ​​of the control parameter within the window to reflect the intensity of parameter fluctuation;

[0069] (2) Control trigger count: Count the number of commands in the window as a measure of control density;

[0070] (3) Average response delay: Calculate the average response delay value of all instructions in the window to evaluate the response performance.

[0071] All statistical results are integrated into a set of three-dimensional indicators, constituting the channel control behavior vector for the current sliding window. As the window slides, a series of continuous control behavior vector sequences are obtained, fully covering the control dynamic characteristics of the channel within the disturbance period. The above sliding window calculation process is performed on each channel group to form a set of channel-level behavior vectors. Finally, the behavior vectors of each channel are concatenated on the time axis according to their timestamp order to construct a two-dimensional time-channel behavior trajectory matrix.

[0072] In the disturbance mapping module, disturbance factor data is collected, and a mapping relationship between disturbance factors and behavioral trajectories is established.

[0073] In the operating environment of new energy equipment, external disturbance factors widely exist that affect the equipment's operating status. These disturbance factors are collected in real time through an Internet of Things (IoT) sensing network and can be directly mapped into the equipment's operation and adjustment logic. To ensure the accuracy of disturbance analysis, the functional relationship between the equipment and the disturbance factors must first be clarified. For example, for photovoltaic inverter equipment, it is mainly affected by environmental factors such as solar irradiance, external air temperature, and module surface temperature; for wind power generation equipment, disturbance factors such as wind speed, wind direction, and air density need to be considered. Therefore, a one-to-one selection criterion for disturbance factors should be established according to the equipment type, and factor signal sources that meet the conditions should be screened out in the sensing network. In engineering implementation, data from environmental sensing units deployed on-site, including temperature and humidity sensors, irradiance acquisition devices, and anemometers, are used to filter out disturbance factor data segments consistent with the equipment's operating cycle through synchronized timestamps. The extracted data should meet basic requirements such as complete temporal continuity, consistent sampling frequency, and unified unit specifications. For example, to monitor the operation of inverters for a group of photovoltaic arrays, irradiance (unit: W / m²) and temperature (unit: °C) data are extracted daily from 10:00 to 16:30, with a sampling interval of 5 seconds, to form a complete disturbance factor dataset.

[0074] After completing the data collection of disturbance factors, time series analysis needs to be performed on each disturbance factor to identify the abrupt change points of the disturbance signal and divide the entire disturbance process into multiple physically interpretable stages. First, the first derivative of the disturbance factor sequence is calculated to represent the instantaneous rate of change of the disturbance factor. This rate of change can be calculated by dividing the difference between adjacent sampling points by the sampling interval, reflecting the disturbance intensity and fluctuation trend. In actual implementation, taking irradiance disturbance as an example, the sampling sequence is processed by sliding window differencing to generate a rate of change sequence. Then, a disturbance amplitude threshold is introduced to judge the rate of change sequence. The threshold setting should be based on historical operating data statistics of the equipment and combined with weather forecasts. For example, if the system identifies that the irradiance changes by more than 150 W / m² within two consecutive sampling periods and the duration exceeds 30 seconds, the current point can be determined as the disturbance initiation point; when the rate of change drops below 50 W / m² and the fluctuation amplitude remains less than 10%, it is determined as the disturbance mitigation point.

[0075] The time interval between the disturbance initiation point and the mitigation point constitutes the complete disturbance process. This process can be further divided into three stages based on the trend of change: the initiation period (the stage where the disturbance change rate rises or falls significantly), the stabilization period (the disturbance remains near its extreme value or the fluctuation trend tends to be stable), and the decay period (the amplitude of the disturbance change gradually decreases and tends to stabilize). For each stage, its start and end times, the range of disturbance factor values, the fluctuation range of the change rate, and the duration should be recorded.

[0076] After completing the disturbance stage labeling, the disturbance labels are accurately mapped to the time axis of the control behavior trajectory matrix of the new energy equipment, realizing the linkage between the disturbance environment and the control response. The time axis index information of the control behavior trajectory matrix is ​​called. This time axis should originate from the execution timestamp generated during the control command acquisition stage, typically possessing precise second- or millisecond-level time resolution. By matching time intervals with control command timestamps, the location and mapping of disturbance stages on the behavior trajectory matrix can be achieved. Specifically, the timestamp of each control command is judged; if it falls within a certain disturbance stage interval, the matrix position of that control behavior is labeled with the corresponding disturbance label, such as "introduction period," "stabilization period," or "decay period." To avoid label overlap or gaps, it should be ensured that the disturbance stage labels have non-overlapping time coverage, and each control behavior should be marked as a unique disturbance stage. For situations where multiple disturbance factors coexist, a disturbance label overlay method should be used, storing the disturbance stage labels of each channel in a combined label format.

[0077] After mapping the disturbance stage labels to the behavior trajectory matrix, corresponding behavior trajectory segments are extracted based on the time periods covered by each disturbance stage, and their internal control characteristics are statistically analyzed and quantified. The execution targets are the time-aligned disturbance labels and control trajectory matrix. First, the trajectory matrix is ​​segmented according to the start and end time intervals of each disturbance stage. Each trajectory segment is a set of instruction sequences for all control channels in the matrix within that time period, including control parameter values, channel numbers, trigger times, and feedback response times. After extracting the trajectory segments, the control parameter variation characteristics need to be statistically analyzed. This includes indicators such as the maximum, minimum, average, standard deviation, and average magnitude of parameter increments. For example, during the disturbance induction period, if the parameter of a certain control channel continuously increases from 0.6 to 1.2, its variation range can be recorded as 0.6, the maximum growth rate as 0.3 per second, and the standard deviation within the statistical period as 0.18. All channels are processed in the same way to form a complete control parameter variation statistics table. Subsequently, the trigger frequency of each control channel during the corresponding disturbance phase is statistically analyzed, that is, the number of times a control command is executed by that channel per unit time. The statistical method can employ a fixed-duration sliding window approach, for example, with a window size of 5 seconds and a sliding step of 1 second, counting the number of triggers within each window. The final output includes the average, peak, and volatility of the trigger frequency during the disturbance phase. This information can be used to measure the control activity of the device during disturbances and reflect the sensitivity of the system response. Finally, for the statistical processing of feedback response delay, the difference between the feedback time and the issuance time of the control command needs to be extracted, and the average, maximum, and trend of the response delay for each channel during the disturbance phase needs to be calculated.

[0078] After extracting statistical features of various control behaviors during the disturbance phase, this information is structured, organized, and stored uniformly to construct a disturbance response dataset for subsequent modeling and analysis. The basic unit of this dataset is a "disturbance response sample," with each sample corresponding to a disturbance phase and including multiple fields. The disturbance phase label serves as the core classification information for the sample, containing the disturbance type (e.g., wind speed, irradiance, temperature), disturbance phase (introduction, stabilization, decay), and corresponding time interval. Simultaneously, the rate of change of the disturbance factor is incorporated into the sample structure. This rate of change includes statistical descriptive terms such as the mean of the first derivative of the disturbance factor, maximum amplitude of change, direction of change, and location of abrupt change points, reflecting the severity and duration of the disturbance environment. The third part is a set of statistical features for behavioral trajectory segments, covering all statistical indicators extracted in the preceding steps, such as control parameter fluctuations, trigger frequency, and feedback response delay. All metrics need to be structured and encapsulated. Each channel should be constructed with a feature substructure containing standard fields, including channel number, average parameter variation, parameter fluctuation range, average trigger frequency, maximum trigger rate, average response delay, maximum response delay, and delay change trend.

[0079] In the offset recognition module, based on the tail trajectory state after the disturbance phase ends, it identifies whether the tail trajectory deviates from the initial trajectory and records the offset path of the corresponding control channel.

[0080] When identifying whether the control behavior of new energy equipment deviates due to disturbances, the trajectory change characteristics at different stages within the disturbance response cycle are extracted and analyzed. First, the end time point of the disturbance decay period is determined from the constructed time-channel two-dimensional behavior trajectory matrix. This point signifies that the impact of the external disturbance has been largely eliminated, and the equipment control behavior has entered a stable phase after the disturbance. The specific end time point of the disturbance decay period can be identified by the first derivative of the disturbance factor data curve approaching zero and maintaining a fluctuation amplitude below a set threshold within a certain time interval. After identification, a fixed-length segment of behavior trajectory data is extracted from this time point. The time length can be set according to the characteristics of the equipment control cycle; the default setting is 20 minutes to ensure that the extracted tail segment of the trajectory fully covers the behavioral characteristics of the initial stable operating phase after the disturbance.

[0081] After extracting the tail segment trajectory, a detailed analysis of the multi-channel control behavior contained within this trajectory is performed. Each control channel in the tail segment trajectory is traversed, extracting the control parameter output sequence, control command execution order, and feedback response time for each channel. Control parameters include, but are not limited to, target values ​​of specific controllable objects such as temperature setpoints, voltage adjustments, and power output thresholds. The control command execution order can be directly obtained through the column position index of the trajectory matrix. The feedback response delay is obtained by subtracting the issuance time of the control command from the response time of each control action. These three dimensions of indicators construct the basic behavioral characteristics for each channel. For the statistical analysis of variation amplitude, the difference between the maximum and minimum values ​​of the control parameters in the tail segment is used, and the standard deviation is calculated. The activation order is statistically based on the time position of the channel's first command response in the tail segment trajectory, sorted according to the time axis. Response consistency is measured by the standard deviation of the feedback delay value; a smaller standard deviation indicates stable feedback behavior, while a larger standard deviation indicates fluctuating feedback behavior. The above three sets of statistical indicators are concatenated into a structured feature vector, forming a tail segment trajectory state vector matrix, where each row corresponds to the complete state characteristics of a channel. This state vector matrix provides the basic data input for subsequent comparison with the initial trajectory state, ensuring that offset recognition does not rely on a single indicator, but integrates multi-dimensional control behavior features to comprehensively reflect the changing trend of the control path.

[0082] To identify whether the tail segment trajectory has deviated from the initial control state, an effective reference segment needs to be established. Specifically, by tracing back the time interval before the disturbance initiation point that is the same length as the tail segment trajectory, a historical trajectory segment of the corresponding length in the behavior trajectory matrix is ​​extracted as the initial trajectory. The timestamp of the disturbance initiation point is obtained by locating the point when the rate of change of the disturbance factor exceeds a set threshold; the trajectory before this point is considered the standard operating condition control process before the disturbance influence. To avoid affecting alignment accuracy due to inconsistent sampling frequencies or differences in local response rates, a trajectory compression alignment method is used for time normalization processing. Trajectory compression alignment refers to using methods such as resampling or interpolation to ensure that two trajectory segments have consistent time segmentation granularity in the same channel dimension. After alignment, the consistency of control parameter changes, activation order, and feedback delay is extracted using the same method as for the tail segment, forming the initial trajectory state vector. Next, the tail segment state vector and the initial state vector are subjected to a one-to-one difference operation on a channel-by-channel basis to generate a multi-dimensional channel vector difference matrix. The difference in each channel dimension represents the degree of deviation of the behavior state of that channel after the disturbance ends compared to before the disturbance. To avoid outliers interfering with the identification process, it is recommended to perform extreme value removal (e.g., removing the maximum and minimum 5%) and normalization on the difference matrix.

[0083] After calculating the multi-dimensional channel differences between the tail trajectory state vector and the initial behavior trajectory state vector, stability tests must be performed on these difference results to determine whether the equipment control behavior has deviated significantly. First, a reference index range for stability judgment is set for each channel dimension. This range consists of the mean range obtained from historical stable operating samples and the allowable variance range. Stability difference samples under normal equipment operation conditions, without obvious system adjustment intentions, are extracted from the behavior trajectories before a large number of disturbances to establish a true stability index baseline. For example, in a wind power system, the behavior parameters of the control channels can be extracted from steady-state operating segments under multiple wind speed environments, and their mean and standard deviation can be statistically analyzed as a benchmark.

[0084] The difference sequence between the tail trajectory state vector and the initial trajectory state vector is input into the stability test program. For each channel dimension, the difference sequence is divided into several time periods, and the mean and variance of each period are calculated. These statistics are compared with a preset stability index range. If the mean exceeds the stable mean range or the variance significantly exceeds the allowable fluctuation range in any time period, the channel dimension is marked as an unstable channel, and the tail trajectory state corresponding to that channel is determined to have deviated from the original behavioral trajectory feature space. To avoid individual short-term fluctuations affecting the overall judgment, a continuous unstable segment count detection is performed; that is, multiple consecutive detection segments must exceed the stable range within a certain time period to constitute the final offset determination. Finally, if the channel exhibits the above-mentioned stability failure, the overall tail trajectory is marked as having a trajectory feature offset state. The current number of offset channels is checked to see if it reaches the threshold requirement. The judgment results of each channel are integrated to form an offset dimension judgment set. The channel number threshold is set according to the control architecture characteristics and operating experience of the equipment; the default is one. For example, in wind turbine equipment, the number of offset channels may be set to three or more as the standard, while in photovoltaic inverters, it may be set to five or more. If the current number of offset channels does not reach the threshold, the system will maintain the tail segment trajectory as a "non-offset path"; otherwise, the tail segment trajectory will be confirmed as an "offset path". After confirming the tail segment trajectory as an offset path, the specific directional characteristics of the offset channels are further recorded and analyzed. To this end, the trend of the difference sequence direction in each offset channel dimension is extracted. This trend can be determined by judging whether the overall sign (positive or negative) of the difference remains consistent after the disturbance ends, and whether there is a continuous increasing or decreasing trend. If the control parameter value corresponding to a certain channel in the tail segment trajectory continues to increase compared to the initial state, the offset direction of that channel is "positive offset"; if it continues to decrease, it is "negative offset"; if there is a sharp jump or irregular fluctuation, it is recorded as "unstable direction".

[0085] In the inertial recognition module, when a trajectory feature offset is detected, the multi-channel evolution trend vector of the tail trajectory is extracted to construct a channel-level inertial trend structure.

[0086] After determining the trajectory deviation and marking the deviation path, the evolution of control behavior is further analyzed in depth to extract the complete evolution of control parameters for each channel in the deviation path within the time interval after the disturbance. Specifically, the set of control channels in the deviation state in the tail segment of the trajectory is first identified, and the control command records for each channel within a fixed time period after the disturbance ends are located. The parameter value and timestamp corresponding to each control command together constitute the basic data points of the control behavior. Subsequently, for each control channel, a continuous control parameter change curve is constructed with time as the horizontal axis and control parameter value as the vertical axis. Each curve has a uniform time resolution and sample density. Unlike traditional methods that extract parameter snapshots at once, this process requires complete tracking of the continuous control behavior throughout the entire post-disturbance process to ensure that every subtle change is accurately captured.

[0087] After obtaining the control parameter change curves for each channel, trend modeling is performed. The core of trend modeling is to identify the direction and speed of change for each curve, rather than simply extracting the final parameter values. On each control parameter change curve, based on a sliding sampling method with equal intervals, the rate of change of parameter values ​​between adjacent time periods is calculated. This rate of change is the local slope of the curve in the current time period, a key indicator characterizing the evolution trend. The sign and magnitude of the slope reflect the speed at which the parameter value rises or falls. The system constructs a trend sequence from all these slope values ​​to reflect the trend characteristics of the entire curve. For example, if a channel parameter continues to rise after a disturbance ends, its corresponding trend sequence will show a continuous positive slope; while if another channel first falls and then rises, its trend sequence will show a transition pattern from a negative slope to a positive slope. After completing the trend sequence construction, the trend samples are further grouped. The grouping is based on the disturbance stage label, namely the disturbance introduction period, stabilization period, and decay period clearly defined in the aforementioned processing. Each trend sequence will be categorized into the trend set within the corresponding stage based on the disturbance stage of its offset path. The purpose of this is to differentiate and manage control response behavior under different disturbance modes, so that trend sequences under the same disturbance condition are grouped together, forming a systematic correlation between disturbance type and control trend.

[0088] After trend grouping, common evolutionary patterns of channel-level control behaviors are explored, and cluster analysis is performed on the trend sequences within the group. The core objective of the clustering method is to group trend sequences with similar morphologies into the same category, using a representative trend in each category as a central template to construct a trend classification space with inductive capabilities. The clustering method does not use the traditional mean distance method, but instead employs a curvature similarity-based strategy. Specifically, the first and second-order change features of each trend sequence are extracted to construct a corresponding "trend curvature vector." This vector reflects the rate of change and bending characteristics of the curve, serving as the core basis for measuring whether two trends are similar in morphology. Subsequently, a clustering algorithm is used to cluster all trend curvature vectors, selecting the most representative trend sequence in each category as the trend central template. The trend central template consists of trend sequences with stable curvature structures and clear channel responses, effectively representing the evolutionary path of a certain type of control behavior. After constructing the trend classification space and extracting multiple trend central templates, the tail segment trajectory to be identified is assessed within this trend classification space to determine whether multi-channel control behaviors converge to a particular trend template. The key to this process is not whether the single-channel trend matching is similar, but whether multiple control channels exhibit consistent behavioral inertia in the tail trajectory, thereby forming a identifiable inertial trend structure.

[0089] The system acquires a set of tail-segment trajectories with completed disturbance segmentation and trajectory offset identification from the behavioral trajectory matrices of multiple new energy devices (or multiple disturbance cycles of the same device). For each tail-segment trajectory, the system re-extracts the trend sequence within its control channel set. These trend sequences have previously been constructed as slope sequences with varying parameters by channel. At this stage, the system reformatts the trend sequence of each channel into a feature vector format that matches the trend classification space. This is based on a curvature expression of uniform length and weights key inflection points according to category weights, ensuring sufficient resolution and directional sensitivity when mapping the trend pattern to the trend space. Subsequently, the trend vectors of each channel in each tail-segment trajectory are sequentially input into the trend classification space for mapping. Specifically, each trend vector is compared with all central templates in the classification space using curvature similarity calculation. This calculation uses the local slope difference, global turning direction difference, and trend stability coefficient between vectors as evaluation criteria, ultimately outputting a trend matching score. The higher the score, the stronger the convergence of the current trend to the central template. Each trend vector is assigned to the class corresponding to the center template with the highest matching score, and its matching confidence is recorded.

[0090] After completing the trend mapping for all channels, the system will perform cross-analysis on the matching results of all channels in each tail segment trajectory to determine whether there is a "multi-channel trend consistency convergence" phenomenon. If, in the same tail segment trajectory, the trend vectors of at least three or more control channels are assigned to the same trend template, and their matching scores are all higher than the set recognition threshold (e.g., above 0.85), the system will consider that the current tail segment trajectory has exhibited a "multi-channel consistent trend convergence" phenomenon. This phenomenon indicates that the equipment control system has formed a fixed adjustment mode across channels after the disturbance is removed, i.e., a behavioral inertia trend.

[0091] In the model training module, the trajectory trend recognition model is trained, and the inertial trend structure and the control strategy output are used together to form feature inputs to determine whether the control strategy has a reference memory offset.

[0092] Based on the identified behavioral inertia trend structure, key trend evolution features are systematically extracted as input features for subsequent model training. Specifically, the control channel set, path convergence period, and slope of the control parameter change curve are extracted as main feature parameters from each tail trajectory marked as having an inertia trend. The control channel set represents the channel number and identifier that actually participates in trend convergence within the inertia trend; instead of using all channels, training samples are established only for the key channels that generate trend convergence. The path convergence period is defined as the time length required for the control parameter change trend in the tail trajectory to stably point towards a certain trend center template. It is calculated as follows: starting from the beginning of the tail trajectory, the system performs time evolution analysis on the matching confidence of the trend vector of each channel in the trend classification space. When the built-in confidence continuously increases and stabilizes near a certain center template over multiple consecutive time windows (e.g., confidence greater than 0.85 and continuous duration exceeding the set window length), the convergence process is considered complete. This period length not only reflects the response speed of trend convergence but also helps determine the stable adjustment speed of the control system after disturbances. Extracting the slope of the control parameter variation curve is one of the key features of the input. An original curve depicting the control parameter changing over time is constructed on the tail segment trajectory of each channel, and the first derivative sequence of its curvature change process is extracted using a local window fitting method. This slope sequence not only describes the rate of parameter change but also reflects the sensitivity and inertia of the system during adjustment. Each channel outputs multiple quantified features, such as average slope, maximum slope, and slope variation coefficient, as part of the final input vector.

[0093] After completing the feature extraction, the system integrates the above content with the perturbation label information, perturbation factor change trends, and automatic control unit policy output parameters in the corresponding perturbation response dataset to generate training samples in a unified format. The training sample structure must ensure consistent dimensions and standardized fields to avoid dimensional inconsistencies or data drift during model training. The sample labels are derived from the system's initial judgment of whether the tail trajectory is in a "baseline offset state." Samples identified as having a baseline offset are marked as positive samples, while trajectories without obvious trend convergence or policy offset are marked as negative samples, ensuring the model's discriminative ability. The baseline offset state is determined based on the path offset direction; samples with a "positive offset" or "negative offset" direction are considered positive samples, while those with an "uncertain direction" are considered negative samples.

[0094] The training model employs a classification model based on a temporal residual aggregation mechanism. This mechanism not only classifies and judges trend features at a single moment but also sequentially inputs the trend evolution feature sequence of each time window within the tail trajectory, achieving a global judgment of the overall trajectory trend through the accumulation and aggregation of temporal residuals. During training, the model learns the static boundaries between features and simultaneously captures the dynamic evolution patterns of control behavior trends during disturbance response. This ensures the model possesses temporal correlation recognition capabilities and trend classification accuracy, making it suitable for the feature structure of "slow slippage without instantaneous abrupt changes in path deviation" in energy control behaviors. To improve the model's generalization ability, the system employs a multi-round cross-training mechanism during the training phase. All sample sets are divided into multiple subsets, which are alternately used as validation and training sets. After each round of training, the system evaluates the model's accuracy, recall, and confidence in deviation judgment, and selects the optimal parameter configuration from multiple rounds for final model freezing. This ensures that the model is not simply a memorization of the training set but truly possesses the ability to recognize trends and judge deviation tendencies.

[0095] After training, the model will be deployed in a real-time recognition system. It will then perform baseline memory offset judgment on the tail segment trajectory features generated during real-time operation. Based on the previously completed data cleaning and channel structuring, the model ensures that the tail segment trajectory contains a stable convergence trend and sufficient control behavior information. Subsequently, feature parameters of the tail segment trajectory across all control channel dimensions are extracted, including but not limited to the slope information of the control parameter change curves, channel activation order, inter-channel collaborative change patterns, and the convergence period of the evolution path. These features will be used as a unified vector structure and input into the trajectory trend recognition model. During recognition, the model maps the tail segment trajectory vector to the trend template space of each offset path and calculates the degree of fit between it and each trend center template. This degree of fit is typically expressed as a matching score or distance index, representing the convergence confidence of the tail segment trajectory in the corresponding trend path. To provide a structured basis for offset judgment, the model outputs the fitting results of each path as normalized confidence labels. Each label corresponds to the matching confidence of a trend path, with the highest confidence value representing the most likely behavioral evolution category to which the current tail segment trajectory belongs.

[0096] In this embodiment, the automatic control unit refers to the key component in the new energy equipment that undertakes the actual control decision-making and execution functions. It is usually deployed in the core control loop and analyzes the current state of the equipment in real time through pre-loaded control logic or strategy algorithms, and outputs parameter commands for controlling the execution of various components. During the operation of the new energy equipment, the automatic control unit does not passively accept external adjustment signals, but automatically analyzes the collected equipment state parameters, external disturbance information, and historical control records based on its own internally embedded operating strategy, and generates new control decisions accordingly.

[0097] The control parameters output by this automatic control unit are called strategy output parameters. These parameters have the following significant characteristics: First, strategy output parameters do not originate directly from human-machine interface input or external adjustment commands, but are automatically calculated and generated by a preset control algorithm within the control unit. Second, these parameters are typically calculated based on multi-channel input signals of the equipment status, such as temperature, current, voltage, speed, and load pressure, through an embedded control model (e.g., PID control logic, fuzzy control logic, or an adaptive model based on state feedback), thus exhibiting high responsiveness and timeliness. Finally, strategy output parameters not only cover target setpoints but may also include complex structures such as adjustment increments, execution channel selection signals, and constraint boundary parameters.

[0098] In the reset maintenance module, when a baseline offset state is identified, an automatic reset command sequence is constructed by backtracking the adjustment path based on the identification result, and a parameter rollback instruction is sent to the control strategy.

[0099] The trajectory trend recognition model outputs a corresponding offset path recognition result based on the fitting result of the tail trajectory in the trend classification space. This result includes the associated offset path and a confidence label used to quantify the recognition's reliability. This confidence label is generated based on a combination of multiple dimensions of features, such as the convergence of distance, consistency of change direction, and curvature matching degree of the trend vectors of each channel of the tail trajectory within the trend space, and possesses the ability to reflect trend offset. After obtaining this label, the offset confidence of each channel should be compared one by one with a preset confidence threshold. The confidence threshold should be empirically calibrated based on historical training data before deployment, for example, by obtaining a 90% confidence boundary as a steady-state baseline through statistical analysis of samples in a non-offset state, thereby ensuring that offset recognition does not produce false alarms or missed alarms. When the confidence value of a channel exceeds this threshold, the current control state of that channel needs to be marked as a parameter offset state, triggering the subsequent automatic reset adjustment process. This judgment process must be executed channel by channel, taking into account the cross-influence relationships between multiple channels to avoid local fluctuations interfering with the overall recognition judgment.

[0100] Once the offset confidence level of a channel exceeds a threshold, the current control state of that channel must be marked. This status mark should be included in the execution trigger conditions of subsequent parameter backtracking and reset commands to ensure that reset operations are not misused or omitted. Simultaneously, the recorded information of this offset state needs to be structured, including the marked time point, the corresponding disturbance segment identifier, the current control parameter value, the corresponding trend label, and the fitted offset direction, providing complete background data support for subsequent reset path analysis and adjustment action execution. The channel's adjustment command sequence from the disturbance initiation point to the current time is backtracked to extract the compensation path for the correction parameters.

[0101] After marking the parameter offset state, a backtracking operation is immediately performed to analyze the entire adjustment behavior process of the channel from the disturbance initiation point to the current time. The core of this process is to extract the correction path from the adjustment command sequence, that is, the set of adjustment behaviors passively or actively generated by the device in the process of attempting to maintain steady state. In the operation, all adjustment control commands within this time period need to be extracted first, arranged in chronological order, and the parameter change amplitude, feedback delay, and inter-channel interaction effects corresponding to each adjustment command are statistically analyzed. Combined with its control objective, it is determined whether it belongs to an exploratory correction operation. Subsequently, a parameter change curve evolving over time is constructed, which is the original basis of the compensation path. The compensation path contains various types of adjustment behaviors, including short-term oscillating correction, continuous drift adjustment, and phased stabilization process. These change patterns are classified and labeled to facilitate the subsequent extraction of the reset reference point. The establishment of the entire compensation path is essentially a structured reconstruction of the recent control history of the channel, providing real and effective reference data for the design of the reset strategy. Specifically, the direction of change in the feedback control response after each adjustment command is issued is statistically analyzed, and it is determined whether the change shows a trend of returning to the stable range. If the direction of change in the adjusted control parameters is opposite to the direction of the tail-segment trajectory offset, and the parameter values ​​at multiple subsequent time points continue to approach the steady-state baseline before the disturbance initiation point, the command can be marked as an attempt at correction. Furthermore, by connecting the sequence of all commands marked as correction behaviors in chronological order, the channel's correction compensation path can be constructed. This path reflects the tendency of the equipment to automatically return to stability under offset conditions initiated by the control system.

[0102] After the compensation path is established, the instantaneous point representing the most significant deviation of the device from the baseline state is identified. This point is typically represented by the maximum rate of change of the control parameters per unit time, i.e., the maximum slope point or inflection point. To identify this point, the magnitude of the parameter difference change before and after each instant can be calculated based on a sliding window, and the time point corresponding to the maximum change magnitude is selected as the maximum offset gradient point. Before this point, there is usually a relatively stable range of parameters, which can be determined by identifying several consecutive sampling points where the rate of change of parameters is lower than a set stability threshold. This stable parameter value is considered the possible operating baseline state of the channel and is denoted as the proposed reset baseline value. Next, using this baseline value as the target, a reset adjustment command sequence is constructed in conjunction with the current control parameter difference. The sequence must guide the parameters back step by step according to the current strategy logic to avoid control oscillations caused by excessively rapid backtracking. After construction, the command sequence will be sent to the corresponding channel control unit of the control system to force the strategy parameter backtracking and ensure that the control logic returns to the steady-state strategy framework. After the reset is executed, the system needs to continuously monitor the channel behavior to determine whether it has successfully stabilized. If it has not, manual intervention maintenance is required.

[0103] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0104] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0105] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0106] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0107] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0108] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0109] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0110] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0111] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0112] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent operation and maintenance management system for new energy equipment based on the Internet of Things, characterized in that, include: The trajectory extraction module is used to collect the control command sequence during the operation of new energy equipment and construct a two-dimensional behavior trajectory matrix that reflects the timing of control actions. The perturbation mapping module is used to collect perturbation factor data and establish a mapping relationship between perturbation factors and behavioral trajectories. The offset recognition module is used to identify whether the tail trajectory deviates from the initial trajectory based on the tail trajectory state after the disturbance phase ends, and to record the offset path of the corresponding control channel. The inertial recognition module is used to extract the multi-channel evolution trend vector of the tail segment trajectory and construct a channel-level inertial trend structure when trajectory feature offset is detected. The model training module is used to train the trajectory trend recognition model. It combines the inertial trend structure and the control strategy output to form the feature input and determine whether the control strategy has a reference memory offset. The reset maintenance module is used to construct an automatic reset command sequence based on the identification result when the reference offset state is identified, and to send parameter rollback instructions to the control strategy.

2. The intelligent operation and maintenance management system for new energy equipment based on the Internet of Things according to claim 1, characterized in that, The trajectory extraction module collects the control command sequence during the operation of new energy equipment and constructs a two-dimensional behavioral trajectory matrix reflecting the timing of control actions, specifically including: Collect all control commands of new energy equipment within the disturbance response cycle, and extract the control parameters, control channel identifier, feedback response delay and corresponding execution timestamp from each command; The extracted instruction sequence is arranged in chronological order and divided into channel groups according to control channels; Within each channel group, control action segments are divided based on a sliding window with a fixed time width. The magnitude of control parameter changes, the number of control triggers, and the average feedback response delay within each segment are statistically analyzed and integrated into a channel-level control behavior vector. The behavior vectors of each channel are concatenated in chronological order to construct a structured time-channel two-dimensional behavior trajectory matrix.

3. The intelligent operation and maintenance management system for new energy equipment based on the Internet of Things according to claim 1, characterized in that, The disturbance mapping module collects disturbance factor data and establishes a mapping relationship between disturbance factors and behavioral trajectories, specifically including: Disturbance factor data associated with new energy equipment is extracted from the Internet of Things sensing network, and the disturbance factors correspond to the functional types of new energy equipment; Based on the first derivative of the rate of change of the disturbance factor data and the disturbance amplitude threshold, the disturbance initiation point and the disturbance mitigation point are identified, and the disturbance is divided into three stages: initiation period, stable period and decay period according to the duration and change pattern of continuous disturbance. The identified perturbation phase labels are mapped to the time axis of the behavior trajectory matrix according to the time alignment method; Record the parameter changes, control channel trigger frequency, and local statistical characteristics of the behavior trajectory segment corresponding to each disturbance stage; The statistical characteristics of the disturbance stage labels, the change rate of the disturbance factor data, and the behavioral trajectory segments are uniformly stored as a labeled disturbance response dataset.

4. The intelligent operation and maintenance management system for new energy equipment based on the Internet of Things according to claim 1, characterized in that, In the offset recognition module, based on the tail trajectory state after the disturbance phase ends, it identifies whether the tail trajectory deviates from the initial trajectory and records the offset path of the corresponding control channel. Specifically, this includes: Extract a fixed-length segment of the behavior trajectory from the behavior trajectory matrix after the end of the disturbance decay period as the tail trajectory. Extract the control parameters of the tail trajectory in each channel, and statistically analyze the activation order, control parameter variation range and feedback response delay consistency of each channel to construct the tail trajectory state vector; Call the initial behavior trajectory segment with the same length as the tail trajectory before the disturbance import point, extract the initial behavior trajectory state vector using trajectory compression and alignment, and perform multi-dimensional channel vector difference calculation with the tail trajectory state vector. According to the preset stability index range, the mean and variance stability tests are performed on the difference sequence to determine whether the tail trajectory state has deviated from the feature space of the initial behavior trajectory. If so, it is marked as trajectory feature offset. When the tail segment trajectory has trajectory feature offset in at least a set number of channel dimensions, the corresponding tail segment trajectory is marked as an offset path, and the offset channel and offset direction are recorded.

5. The intelligent operation and maintenance management system for new energy equipment based on the Internet of Things according to claim 1, characterized in that, In the inertial recognition module, when a trajectory feature shift is detected, the multi-channel evolution trend vector of the tail segment trajectory is extracted, and a channel-level inertial trend structure is constructed, specifically including: Extract the time evolution curves of the control parameters output for each channel from the offset path that has been marked as trajectory feature offset state; In the control parameter change curves of each channel, an evolution trend sequence is constructed based on the change in curve slope, and the channel evolution trends under different perturbations are grouped according to the perturbation stage label. Clustering methods are used to establish a path change pattern classification set for the trend vector sequence. Curvature similarity is performed on the control parameter curves corresponding to the channels within the group, and trend center templates are extracted to construct a trend classification space. Multiple tail trajectories are mapped in the trend classification space to identify whether they converge to a specific trend center template. If the corresponding convergence path appears in multiple channels at the same time, the convergence path is marked as a behavioral inertia trend structure.

6. The intelligent operation and maintenance management system for new energy equipment based on the Internet of Things according to claim 1, characterized in that, In the model training module, the trajectory trend recognition model is trained, and the inertial trend structure and the control strategy output are used together to form feature inputs. The determination of whether the control strategy has a baseline memory offset specifically includes: The channel and path convergence period and the slope of the control parameter change curve extracted from the behavioral inertia trend structure are used as feature inputs. By combining the disturbance response dataset with the policy output parameters of the automatic control unit, training samples in a unified format are constructed. A classification model construction method based on temporal residual aggregation mechanism is adopted, and the training samples are cross-trained in multiple rounds to establish a trajectory trend recognition model; Input the tail trajectory features to be judged into the trajectory trend recognition model, calculate the degree of fit of the output results to each offset path, and output the confidence label of the potential benchmark memory offset.

7. The intelligent operation and maintenance management system for new energy equipment based on the Internet of Things according to claim 6, characterized in that, The automatic control unit refers to the component in the new energy equipment that automatically executes decision control. Its corresponding strategy output parameters are control parameters automatically generated by the control program or strategy logic embedded in the automatic control unit.

8. The intelligent operation and maintenance management system for new energy equipment based on the Internet of Things according to claim 1, characterized in that, In the reset maintenance module, when a baseline offset state is identified, an automatic reset command sequence is constructed by backtracking the adjustment path based on the identification result, and a parameter rollback instruction is sent to the control strategy, specifically including: Obtain the potential baseline memory offset confidence label output by the trajectory trend recognition model, and perform confidence threshold determination on the corresponding channel; In channels where the confidence level exceeds the preset offset threshold, the current control state of the channel is marked as parameter offset state; The backtracking channel extracts the compensation path of the correction parameters by tracing all the adjustment command sequences from the disturbance introduction point to the current time. Identify the maximum offset gradient point of the correction parameters in the compensation path, use the parameter of the previous stable segment before this point as the proposed reset reference value to construct an automatic reset adjustment command sequence, and issue a reset command to the controlled channel to execute the strategy output parameter rollback.

Images