New energy equipment intelligent operation and maintenance management system based on Internet of Things

Abstract

The invention discloses a new energy equipment intelligent operation and maintenance management system based on the Internet of Things, particularly relates to the field of equipment intelligent operation and maintenance, and is used for solving the problem that equipment operation control strategy offset is difficult to automatically repair in the prior art. A track extraction module is used for collecting control instructions in the operation process, and a two-dimensional behavior track matrix is constructed; establishing a corresponding relation between a disturbance factor and a behavior track through a disturbance mapping module; judging whether the control path is separated from the original operation mode or not after the disturbance is finished through an offset identification module; when track offset is detected, calling an inertial recognition module to extract an evolution trend of a tail section track; a model training module constructs an offset identification model, and judges whether the control system has reference memory offset or not; and finally, the parameter rollback operation is executed through the reset maintenance module, the control state is automatically restored, and the operation and maintenance efficiency and the system stability of the new energy equipment can be remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment intelligent operation and maintenance, more specifically, the present application relates to a new energy equipment intelligent operation and maintenance management system based on Internet of Things. BACKGROUND

[0002] Under the architecture of Internet of Things, new energy equipment such as wind power, photovoltaic and energy storage is widely equipped with environmental adaptive control systems to respond to external disturbances in real time, such as sudden changes in wind speed, shadow blocking, and power grid voltage fluctuations. Common control response mechanisms include: variable pitch adjustment of wind turbines, maximum power point tracking (MPPT) adjustment of photovoltaic inverters, and dynamic charge-discharge scheduling of energy storage systems. These mechanisms are based on a common assumption that environmental disturbances are short-term, transient and can be quickly recovered, so the control system automatically returns to the design baseline state of the equipment 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 continuous sandstorms, continuous rain causing all-day shading, and continuous voltage disturbances causing charge-discharge restrictions. In this case, the adaptive system will perform frequent compensation for a long time, and its short-term adjustment actions will gradually solidify and evolve into behavioral inertia deviation of the control system. Ultimately, the equipment fails to recover to the design operating condition after the disturbance ends, but instead runs in the compensated state, which is mistakenly recorded as a new stable operating baseline. The operation and maintenance platform cannot identify this type of control parameter deviation due to its dependence on set parameter intervals to determine equipment operating conditions, resulting in a series of chain reactions such as hidden performance decline, reduced strategy adaptability, and abnormal decay of overall efficiency.

[0003] This type of problem has the characteristics of high behavioral structure, parameter deviation, and hidden performance, which is different from physical damage and parameter abnormalities, so there is still a lack of effective identification and intervention mechanism in traditional intelligent operation and maintenance strategies. SUMMARY

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

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions: A new energy equipment intelligent operation and maintenance management system based on Internet of Things, comprising: a trajectory extraction module for collecting control instruction sequences during operation of the new energy equipment and constructing a two-dimensional behavior trajectory matrix reflecting the timing of control actions; a disturbance mapping module for collecting disturbance factor data and establishing a mapping relationship between the disturbance factors and the behavior trajectories; An offset identification module is configured to identify whether the tail track deviates from the initial track based on the tail track state after the disturbance phase ends, and record the offset path of the corresponding control channel; An inertia identification module is configured to extract a multi-channel evolution trend vector of the tail track when it is identified that the track feature deviates, and construct a channel-level inertia trend structure; A model training module is configured to train a track trend identification model, and jointly construct a feature input of the inertia trend structure and the control strategy output to determine whether the control strategy has a benchmark memory deviation; A reset maintenance module is configured to construct an automatic reset command sequence according to the identification result when it is identified that the benchmark deviation state exists, and issue a parameter rollback instruction to the control strategy.

[0006] In a preferred embodiment, the track extraction module collects control instruction sequences during the operation of the new energy equipment, and constructs a two-dimensional behavior track matrix reflecting the control action timing, which specifically includes: All control instructions of the new energy equipment in the disturbance response period are collected, and the control parameters, control channel identifiers, feedback response delays and corresponding execution timestamps in each instruction are extracted; The extracted instruction sequences are arranged in chronological order, and divided into channel groups according to the control channels; In each channel group, control action segments are divided based on a fixed time width sliding window, the control parameter change amplitude, control trigger frequency and feedback response average delay in each segment are counted, and integrated into a channel-level control behavior vector; The channel behavior vectors are spliced in chronological order to construct a structured time-channel two-dimensional behavior track matrix.

[0007] In a preferred embodiment, the disturbance mapping module collects disturbance factor data, and establishes a mapping relationship between the disturbance factor and the behavior track, which specifically includes: The disturbance factor data associated with the new energy equipment is extracted from the Internet of Things perception network, and the disturbance factor corresponds to the functional type of the new energy equipment; Based on the first derivative of the disturbance factor data change rate and the disturbance amplitude threshold, the disturbance introduction point and the disturbance relief point are identified, and the stages are labeled as the introduction period, the stable period and the decay period according to the continuous disturbance duration and the change form; The divided disturbance stage labels are mapped to the time axis of the behavior track matrix in time alignment; The local statistical characteristics of the parameter variation, the control channel trigger frequency and the feedback response delay of the behavior track segment corresponding to each disturbance stage are recorded; The disturbance stage label, the disturbance factor data change rate and the statistical characteristics of the behavior track segment are uniformly stored as a labeled disturbance response data set.

[0008] In a preferred embodiment, in the offset identification module, based on the tail track state after the end of the perturbation stage, whether the tail track deviates from the initial track is identified, and the offset path of the corresponding control channel is recorded, which specifically comprises: A fixed length of behavior track segment after the end time point of the perturbation decay period is intercepted from the behavior track matrix as a tail track; The control parameters of the tail track in each channel are extracted, the activation order, the control parameter variation amplitude and the feedback response delay consistency of each channel are counted, and a tail track state vector is constructed; The initial behavior track segment of the same length as the tail track before the perturbation import point is called, the initial behavior track state vector is extracted in a track compression alignment manner, and multi-dimensional channel vector difference calculation is performed with the tail track state vector; According to the preset stability index interval, the mean and variance stability test is performed on the difference sequence, whether the tail track state has deviated from the characteristic space of the initial behavior track is judged, and if so, it is marked as a track feature offset; When there is a track feature offset in at least a set number of channel dimensions in the tail track, the corresponding tail track is marked as an offset path, and the offset channel and the offset direction are recorded.

[0009] In a preferred embodiment, in the inertia identification module, when the track feature offset is identified, a multi-channel evolution trend vector of the tail track is extracted, and a channel-level inertia trend structure is constructed, which specifically comprises: The change curve of the output of each channel control parameter evolving over time is extracted from the offset path marked as the track feature offset state; In the control parameter change curve of each channel, an evolution trend sequence is constructed based on the change of the curve slope, and the channel evolution trends under different perturbations are grouped according to the perturbation stage label; A path change mode classification set is established for the trend vector sequence using clustering method, the curvature similarity of the control parameter curves of the channels in the group is divided, and a trend center template is extracted, and a trend classification space is constructed; A plurality of tail tracks are mapped in the trend classification space, whether they converge to a specific trend center template is identified, and if the corresponding convergence path appears in multiple channels at the same time, the convergence path is marked as a behavior inertia trend structure.

[0010] In a preferred embodiment, in the model training module, the track trend identification model is trained, the inertia trend structure and the control strategy output are jointly constructed as feature input, and whether the control strategy has a benchmark memory offset is determined, which specifically comprises: The channels, path convergence periods and control parameter change curve slopes extracted in the behavior inertia trend structure are taken as feature input; Combine the disturbance response data set with the strategy output parameter of the automatic control unit to construct a training sample in a unified format; Adopt a classification model construction method based on a time sequence residual aggregation mechanism to perform multi-round cross training on the training sample, and establish a trajectory trend identification model; Input the to-be-judged tail section trajectory feature into the trajectory trend identification model, calculate the fitting degree of the output result in each offset path, and output a confidence label of the potential reference memory offset.

[0011] In a preferred embodiment, the automatic control unit refers to a component for automatically executing decision control in a new energy equipment, and the corresponding strategy output parameter is a control parameter automatically generated by a control program or strategy logic embedded in the automatic control unit.

[0012] In a preferred embodiment of the reset maintenance module, when the reference offset state is identified, an automatic reset command sequence is constructed according to the identification result, and a parameter rollback instruction is issued to the control strategy, specifically comprising: Obtain the potential reference memory offset confidence label output by the trajectory trend identification model, and perform confidence threshold determination on the corresponding channel; Mark the current control state of the channel as a parameter offset state in the channel whose confidence exceeds the preset offset threshold; Backtrack all adjustment instruction sequences in the channel from the disturbance introduction point to the current time, and extract a compensation path of the correction parameter; Identify the maximum offset gradient point of the correction parameter in the compensation path, take the parameter of the previous stable section as a reset reference value to construct an automatic reset adjustment command sequence, and issue a reset instruction to the controlled channel to execute the strategy output parameter rollback.

[0013] The technical effects and advantages of the new energy equipment intelligent operation and maintenance management system based on the Internet of Things are as follows: The application reflects the control behavior characteristics of the new energy equipment in the disturbance response period comprehensively by constructing a time-channel two-dimensional behavior trajectory matrix through a trajectory extraction module, realizes accurate labeling of the disturbance stage and positioning of the behavior influence range by combining a disturbance mapping module to establish the time sequence association of the disturbance factor and the behavior trajectory. An offset identification module automatically identifies the multi-channel state difference between the tail trajectory and the initial trajectory after the disturbance decays, forming a structured judgment of the control behavior characteristic drift. Further, an inertia identification module extracts the multi-channel evolution trend, establishes a trend classification space and an inertia trend structure, and improves the identification ability of the adjustment solidification behavior. A model training module fuses the control strategy output and the trend structure to construct a multi-dimensional feature input, trains a trajectory trend identification model to realize intelligent discrimination of the benchmark memory offset. Finally, a reset maintenance module traces the adjustment path based on the trajectory trend identification model result, accurately identifies the parameter offset position and constructs an automatic reset command sequence, ensuring the pertinence and stability of the control strategy parameter rollback process. The overall scheme breaks through the limitations of the traditional static parameter comparison method, realizes dynamic identification of the disturbance behavior offset, and realizes intelligent closed-loop control of the parameter reset. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A structure schematic diagram of a new energy equipment intelligent operation and maintenance management system based on the Internet of Things is given. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0016] Embodiment 1, Figure 1 A new energy equipment intelligent operation and maintenance management system based on the Internet of Things is given, which comprises: A trajectory extraction module is used to collect the control instruction sequence during the operation of the new energy equipment, and a two-dimensional behavior trajectory matrix reflecting the time sequence of the control action is constructed. A disturbance mapping module is used to collect disturbance factor data and establish the mapping relationship between the disturbance factor and the behavior trajectory. An offset identification module is used to identify whether the tail trajectory deviates from the initial trajectory based on the tail trajectory state after the end of the disturbance stage, and record the offset path of the corresponding control channel. An inertia identification module is used to extract a multi-channel evolution trend vector of the tail trajectory when the trajectory characteristic offset is identified, and construct a channel-level inertia trend structure. The model training module is configured to train the trajectory trend identification model, and the inertia trend structure and the control strategy output are jointly used as feature input to determine whether the control strategy has a benchmark memory deviation. The reset maintenance module is configured to automatically reset the command sequence according to the identification result when the benchmark deviation state is identified, and issue a parameter rollback instruction to the control strategy.

[0017] In the trajectory extraction module, the control instruction sequence during the operation of the new energy equipment is collected, and a two-dimensional behavior trajectory matrix reflecting the control action timing is constructed.

[0018] After the new energy equipment enters the disturbance response period, all control instructions issued by the equipment control layer are recorded, and the key structure fields of each instruction are specified. The control instruction is usually composed of execution identification, target channel number, control parameter value, control issuing time and corresponding feedback response time. In the specific implementation process, the control data acquisition unit first records the instruction of the device controller, extracts continuous original control data from it, and converts it into a structured instruction record sequence.

[0019] Each instruction needs to strictly extract the following field information: (1) Control parameter: indicates the set value for adjustment at this moment, such as output power set value, voltage control value, inverter frequency set value, etc. (2) Control channel identification: specifies which execution component the parameter is applied to, such as motor control unit, inverter power module, temperature control valve, etc., to ensure the accuracy of subsequent channel level grouping; (3) Feedback response delay: defined as the time interval between the issuance of the control instruction and the monitoring of the corresponding feedback signal change, which can be set to millisecond level. The response delay reflects the coupling efficiency between control, execution and feedback; (4) Execution timestamp: records the accurate time point of issuing the control instruction, which is used for subsequent time sequence reconstruction and sliding window positioning.

[0020] Based on the completed structured sequence of control instructions, further unified sorting is performed according to the time axis, and grouping processing is performed according to the control channel to form independent channel behavior data flow. In the specific operation, first, all control instructions are arranged in ascending order according to the execution timestamp field, the complete instruction timing chain is reconstructed, and the time continuity and logical consistency of the control behavior are ensured.

[0021] After the time series reconstruction, all control instructions are divided into their corresponding channel groups according to the control channel identification field. Each channel group represents the complete instruction behavior of the device on the control path, including control value adjustment, trigger frequency, response delay, and other characteristic sequences. For example, for a multi-channel controlled new energy inverter device, its control channels may include output voltage adjustment, frequency conversion control, temperature control fan control, etc. The system will divide the control instructions into independent groups such as channel A, channel B, channel C, etc. according to the channel number.

[0022] After completing the channel grouping, in order to extract control mode features with time locality and behavior integrity from continuous control instructions in each channel, a sliding window mechanism is introduced in each channel group. The sliding window is set to a fixed time width, measured in milliseconds or seconds, and the window width is recommended to be selected within 1-2 times of the typical device response frequency. For example, for a device with a control frequency of 1 Hz, a sliding window of 2 seconds and a window step of 1 second can be set to ensure time sequence continuity and detail coverage.

[0023] In each sliding window, the system extracts all control instructions within that time period and sequentially calculates the following behavior indicators: (1) Control parameter variation amplitude: calculate the difference between the maximum and minimum values of the control parameter in the window, reflecting the parameter fluctuation intensity; (2) Control trigger frequency: count the number of instructions in the window as a measure of control density; (3) Feedback response average delay: take the average of the response delay values of all instructions in the window to evaluate the response performance.

[0024] All statistical results are integrated into a set of three-dimensional indicators to form the channel control behavior vector of the current sliding window. With the window sliding, a series of continuous control behavior vector sequences are obtained, which completely cover the control dynamic characteristics of the channel within the disturbance period. The above sliding window calculation process is performed for each channel group to form a set of channel-level behavior vectors. Finally, the channel behavior vectors are spliced in time sequence according to their time stamps on the time axis to construct a two-dimensional time-channel behavior trajectory matrix.

[0025] In the disturbance mapping module, disturbance factor data is collected to establish a mapping relationship between the disturbance factor and the behavior trajectory.

[0026] In the operating environment of new energy equipment, external disturbance factors that affect the operating state of the equipment are widely present. These disturbance factors are collected in real time through an Internet of Things sensing network and can be directly mapped into the operating adjustment logic of the equipment. In order to ensure the accuracy of the disturbance analysis, the functional association between the equipment and the disturbance factors is first determined. For example, for a photovoltaic inverter device, it is mainly affected by environmental factors such as solar irradiance, external air temperature, and component surface temperature; for a wind power generation device, it needs to pay attention to disturbance factors such as wind speed, wind direction, and air density. Therefore, a one-to-one disturbance factor selection standard should be established according to the type of the equipment, and the factor signal source that meets the condition should be screened out in the sensing network. In engineering implementation, the sensor data of the environmental sensing unit deployed on site is called, including temperature and humidity sensors, irradiance collection devices, anemometers, etc. The disturbance factor data segment consistent with the equipment operation cycle is selected through synchronous time stamp. The extracted data should meet the basic requirements of time continuity, consistent sampling frequency, and unified unit specification. For example, for the operation monitoring of an inverter for a group of photovoltaic arrays, the irradiance (unit: W / m²) and temperature data (unit: °C) during 10:00 to 16:30 every day are extracted, with a sampling interval of 5 seconds, to form a complete disturbance factor data set.

[0027] After completing the disturbance factor data collection, time series analysis is performed on each disturbance factor to identify the mutation point of the disturbance signal and divide the entire disturbance process into multiple stages with physical interpretation. First, the first derivative of the disturbance factor sequence is calculated to represent the instantaneous change rate of the disturbance factor. The change rate 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 change rate sequence is generated by performing sliding window difference processing on the sampling sequence. Then, a disturbance amplitude threshold is introduced to judge the change rate sequence. The threshold should be set based on the statistical data of the historical operation of the equipment and combined with the weather forecast. For example, if the system identifies that the irradiance changes by more than 150 W / m² in two consecutive sampling periods and the duration exceeds 30 seconds, it can be determined that the current point is a disturbance introduction point; when the change rate drops below 50 W / m² and the fluctuation amplitude is less than 10%, it is determined as a disturbance relief point.

[0028] The time period between the disturbance introduction point and the relief point is the complete disturbance process, which can be further divided into three stages according to the change trend: introduction period (stage of significant increase or decrease in disturbance change rate), stable period (disturbance maintained near the extreme value or fluctuation trend tended to be stable), and decay period (disturbance change amplitude gradually decreased and tended to be stable). The start and end time points, disturbance factor value interval, change rate fluctuation range, and duration of each stage should be recorded.

[0029] After completing the disturbance phase label division, the disturbance label is accurately mapped to the control behavior trajectory matrix time axis 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, which should come from the execution timestamp generated in the control instruction collection stage, usually with accurate second-level or millisecond-level time resolution. By matching the time interval and the control instruction timestamp, the positioning and mapping of the disturbance phase on the behavior trajectory matrix can be realized. The specific operation is: judging the timestamp of each control instruction, if it falls within a certain disturbance phase interval, the matrix position of the control behavior is marked with the corresponding disturbance label, such as "import period", "stable period" and "decay period". In order to avoid label overlap or vacancy, it should be ensured that the disturbance phase labels have non-overlapping time coverage, and each control behavior is labeled according to a unique disturbance phase. For the case of coexistence of multiple disturbance factors, the disturbance label superposition method should be used, and the disturbance phase labels of each channel are stored in the form of combined labels.

[0030] After completing the mapping of the disturbance phase label to the behavior trajectory matrix, based on the time period covered by each disturbance phase, the corresponding behavior trajectory segment is extracted, and the control characteristics in it are counted and quantified. The execution object is the disturbance label and the control trajectory matrix that have been time-aligned. First, according to the start and end time interval of each disturbance phase, the trajectory matrix is segmented. Each trajectory segment is a sequence of instructions of all control channels in the matrix within the time period, including control parameter values, channel numbers, trigger time points and feedback response times. After extracting the trajectory segment, the control parameter variation characteristics in it need to be counted. Specifically, it includes the maximum value, minimum value, average value, standard deviation and average amplitude of parameter increment of the control parameter. For example, in the disturbance import period, the parameter of a control channel continuously increases from 0.6 to 1.2, and the variation range is recorded as 0.6, the maximum speed is 0.3 per second, and the standard deviation in the statistical period is 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 in the corresponding disturbance phase is counted, that is, the number of control instruction executions of the channel per unit time is counted. The statistical method can use a fixed time length sliding window method, for example, with a 5-second window and a 1-second sliding step, the number of triggers in each window is counted, and the average, peak and fluctuation rate of the trigger frequency in the disturbance phase are finally output. These information can be used to measure the control activity of the equipment in the disturbance process, 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 issue time is extracted from the control instruction, and the average, maximum and change trend of the response delay of each channel in the disturbance phase are calculated.

[0031] After the extraction of the statistical characteristics of various control behaviors in the disturbance phase, the information is structured and stored uniformly to build a disturbance response dataset that can be used for subsequent modeling and analysis. The basic unit of the dataset is a "disturbance response sample", each sample corresponds to a disturbance phase and includes multiple field contents. The disturbance phase label is used as the core classification information of the sample, which includes the disturbance type (such as wind speed, irradiance, temperature, etc.), the disturbance phase (introduction period, stable period, decay period), and the corresponding time interval. The change rate information of the disturbance factor is also included in the sample structure. The change rate includes the mean value of the first derivative of the disturbance factor, the maximum change amplitude, the change direction, and the position of the abrupt point, which reflects the intensity and duration of the disturbance environment. The third part is the statistical feature set of the behavior trajectory segment, which includes all the statistical indicators extracted in the previous steps, such as control parameter fluctuations, trigger frequency, and feedback response delay. All indicators need to be structured and packaged, and each channel should build a feature substructure containing standard fields, including channel number, parameter variation mean, parameter fluctuation range, average trigger frequency, maximum trigger rate, average response delay, maximum response delay, and delay change trend.

[0032] In the offset identification module, based on the tail trajectory state after the end of the disturbance phase, whether the tail trajectory deviates from the initial trajectory is identified, and the offset path of the corresponding control channel is recorded.

[0033] When identifying whether the control behavior of new energy equipment deviates due to disturbance, the trajectory change characteristics in different stages of the disturbance response period are extracted and analyzed. First, the time node at the end of the disturbance decay period is determined from the already constructed time-channel two-dimensional behavior trajectory matrix, which marks that the external disturbance influence has been basically eliminated and the device control behavior enters the stable stage after the disturbance. The specific end time 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 certain threshold within a certain time interval. After identification, a fixed length of behavior trajectory data is extracted from the time point, and the time length can be set according to the characteristics of the device control period, with a default setting of 20 minutes to ensure that the extracted tail trajectory can completely cover the behavior characteristics of the initial stable stage after the disturbance.

[0034] After the extraction of the tail trajectory, the multi-channel control behavior contained in the trajectory is analyzed in detail. Each control channel in the tail trajectory is traversed to extract the control parameter output sequence, control instruction execution order and feedback response time in each channel. The control parameters include but are not limited to temperature set value, voltage adjustment value, power output threshold and other specific controllable object target values; the control instruction execution order can be directly obtained through the column sequence position index of the trajectory matrix; the feedback response delay is obtained by subtracting the sending time of the control instruction from the response time of each control action. The above three-dimensional indicators construct the basic behavior characteristics of each channel. For the statistical range, the difference between the maximum and minimum values of the control parameters in the tail is measured, and the standard deviation is calculated; the activation sequence statistics is based on the time position of the first instruction response of the channel in the tail trajectory, and is sorted according to the time axis; the response consistency is measured by the standard deviation of the feedback delay value, the smaller the standard deviation, the more stable the feedback behavior of the channel, and the larger the more fluctuant the feedback behavior. The above three groups of statistical indicators are spliced into a structured feature vector to form a tail trajectory state vector matrix, each row of which corresponds to the complete state characteristics of a channel. The state vector matrix provides basic data input for subsequent comparison with the initial trajectory state, ensuring that the deviation identification does not rely on a single indicator, but integrates multi-dimensional control behavior characteristics to fully reflect the control path change trend.

[0035] To identify whether the tail trajectory deviates from the initial control state, an effective reference section needs to be established. Specifically, by tracing back the time interval before the disturbance introduction point and consistent with the length of the tail trajectory, a historical trajectory segment of the same length as the tail is intercepted in the behavior trajectory matrix as the initial trajectory. The disturbance introduction point timestamp is located when the disturbance factor change rate exceeds the set threshold, and the trajectory before this point is considered to be the standard working condition control process before the disturbance. To avoid the influence of inconsistent sampling frequencies or local response rate differences on alignment accuracy, trajectory compression alignment is used for time normalization. Trajectory compression alignment refers to making two trajectory segments have consistent time segmentation granularity under the same channel dimension through resampling or interpolation, etc. After alignment, the control parameter variation, activation sequence and feedback delay consistency are extracted using the same method as the tail to form the initial trajectory state vector. Next, the tail state vector and the initial state vector are executed difference operation one by one according to the channel to generate a multi-dimensional channel vector difference matrix. The difference of each channel dimension represents the deviation degree of the behavior state of the channel after the disturbance ends compared with before the disturbance. To avoid the interference of outliers in the recognition process, it is recommended to perform extreme value removal (such as removing the maximum and minimum 5%) and normalization processing on the difference matrix.

[0036] After the multi-dimensional channel difference calculation of the tail trajectory state vector and the initial behavior trajectory state vector, a stability test must be performed on the difference results to determine whether the device control behavior has deviated significantly. First, a reference index interval for stability discrimination is set for each channel dimension, which is composed of the mean interval and the variance range of the allowed fluctuations obtained from the historical stable running samples. The index parameters are extracted from the behavior trajectory of a large number of pre-disturbance running states, and the stable difference samples are obtained under the condition that there is no obvious system adjustment intention and the device is running normally, to establish a real stable index baseline. For example, in a wind power system, the behavior parameters of the control channel can be extracted from multiple groups of stable running segments under different wind speed environments, and the mean and standard deviation of the changes are calculated as the baseline.

[0037] The difference sequence of the tail trajectory state vector and the initial trajectory state vector is input into the stability test program. In 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 the preset stable index interval. If the mean in any time period exceeds the stable mean range, or the variance significantly exceeds the allowed fluctuation range, the channel dimension is marked as an unstable channel, and it is determined that the tail trajectory state of the channel has deviated from the original behavior trajectory characteristic space. To avoid the influence of individual short-term fluctuations on the overall judgment, continuous unstable segment quantity detection must be performed, i.e., a large number of detection segments within a certain time period must exceed the stable range to constitute the final deviation determination. Finally, if the channel has the above stability failure, the overall tail trajectory is marked as a state with trajectory characteristic deviation. It is checked whether the current number of deviation channels reaches the threshold requirement. The judgment results of each channel are integrated to form a deviation dimension judgment set, and the channel number threshold is set according to the control architecture characteristics and running experience of the device. The default setting is one, for example, in a wind turbine device, the number of deviation channels may be set to three or more, and in a photovoltaic inverter, the number of deviation channels may be set to five or more. If the current number of deviation channels does not reach the threshold, the system will maintain the tail trajectory as a "non-deviation path"; otherwise, the tail trajectory is confirmed as a "deviation path". After confirming that the tail trajectory is a deviation path, the specific direction characteristics of the deviation channels are further recorded and analyzed. For this purpose, the trend of the difference sequence in each deviation channel dimension is extracted. The trend can be completed by judging whether the overall sign (positive or negative) of the difference after the disturbance ends is consistent, whether there is a continuous increasing or decreasing trend. If the control parameter value corresponding to a channel in the tail trajectory continuously increases compared to the initial state, the channel deviation direction is "positive deviation"; if it continuously decreases, it is "negative deviation"; if there is a sharp jump or irregular fluctuation, it is recorded as "direction unstable".

[0038] In the inertia recognition module, when the trajectory characteristic deviation is recognized, the multi-channel evolution trend vector of the tail trajectory is extracted, and a channel-level inertia trend structure is constructed.

[0039] After the trajectory deviation determination and the deviation path marking, the control behavior evolution process is further analyzed, and the evolution process of the control parameters of each channel in the deviation path within the time interval after the disturbance is extracted. The specific operation first confirms the control channel set in the tail trajectory that is in the deviation state, and then locates the control instruction record of these channels within a fixed time period after the disturbance end time point. The parameter value corresponding to each control instruction and the time stamp together constitute the basic data point of the control behavior. Subsequently, in 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 the traditional one-time extraction of parameter snapshots, this process requires complete tracking of the continuous control behavior after the entire disturbance process to ensure that every subtle change can be accurately captured.

[0040] After obtaining the control parameter change curve of each channel, the trend modeling process is performed. The core of the trend modeling is to identify the change direction and speed of each curve, rather than just extracting the final parameter value. On each control parameter change curve, based on the sliding method of equal interval sampling, the parameter value change rate between adjacent time periods is calculated. The change rate is the local slope of the curve at the current period, which is a key indicator of the evolution trend. The sign and value of the slope reflect the speed of the parameter value rising or falling, and the system will construct a trend sequence with all these slope values to reflect the trend characteristics of the entire curve. For example, after the disturbance ends, if a channel parameter continuously rises, its corresponding trend sequence will show a series of positive slopes; if another channel first falls and then rises, its trend sequence will show a transition mode from negative slope to positive slope. After completing the trend sequence construction, the trend samples are further grouped. The grouping basis is the disturbance phase label, which has been clearly divided in the foregoing processing. Each trend sequence will be grouped into the trend set of the corresponding phase according to the disturbance phase of the deviation path it belongs to. The purpose of this is to distinguish and manage the control response behavior under different disturbance modes, so that the trend sequences under the same disturbance condition are gathered together, forming a systematic association between disturbance types and control trends.

[0041] After the trend grouping is completed, the common evolution pattern of the multi-channel control behavior is mined, and the clustering analysis is performed on the trend sequence in the grouped trend. The core goal of the clustering method is to classify the trend sequences with similar shapes into the same class, and to construct a trend classification space with inductive ability by taking the representative trend in each class as the center template. The clustering method does not use the traditional mean distance method, but uses a curvature similarity-based strategy. Specifically, the first-order and second-order change characteristics of each trend sequence are extracted to construct the corresponding "trend curvature vector", which is used to reflect the change speed and bending characteristics of the curve and is the core basis for measuring whether two trends are similar in shape. Then, the clustering algorithm is used to cluster and divide all trend curvature vectors, and the most representative trend sequence in each category is selected as the trend center template. The trend center template is composed of trend sequences with stable curvature structure and clear channel response, which can effectively represent the evolution path of a certain type of control behavior. After the trend classification space is constructed and multiple trend center templates are extracted, the tail trajectory to be identified is judged whether the multi-channel control behavior converges to a certain trend template in the trend classification space. The key of this process is not whether the single-channel trend matching is similar, but whether multiple control channels show consistent behavioral inertia in the tail trajectory, thereby forming a determinable inertia trend structure.

[0042] In the behavior trajectory matrix of multiple new energy devices (or multiple disturbance periods of the same device), the tail trajectory set of the completed disturbance segmentation and trajectory offset recognition is obtained. For each tail trajectory, the system will re-extract the trend sequence in the control channel set, and these trend sequences have been constructed as slope sequences of parameter changes before. In this stage, the system re-formats the trend sequence of each channel into a feature vector format that matches the trend classification space, that is, based on the curvature expression of a uniform length, and weights the key inflection points according to the category weight, so that the trend shape has sufficient resolution and direction sensitivity when mapping to the trend space. Then, the trend vectors of each channel in each tail trajectory are input into the trend classification space for mapping. Specifically, each trend vector will perform curvature similarity calculation with all center templates in the classification space, and the calculation takes the local slope difference, global turning direction difference and trend stability coefficient between the vectors as the basis for judgment, and finally outputs a trend matching score. The higher the score, the stronger the convergence of the current trend vector to the center template. Each trend vector will be assigned to the class corresponding to the center template with the highest matching score, and its matching confidence will be recorded.

[0043] After all the channel trend mapping is completed, the system will cross-analyze the matching results of all channels in each tail trajectory to determine whether there is a "multi-channel trend consistency convergence" phenomenon. If in the same tail trajectory, the trend vectors of at least three and more control channels are assigned to the same trend template, and the matching scores are all higher than the set recognition threshold (for example, above 0.85), the system will consider that the current tail trajectory has appeared "multi-channel consistent trend convergence" phenomenon. This phenomenon indicates that the device control system has formed a cross-channel fixed adjustment mode after the disturbance is removed, that is, the behavior inertia trend.

[0044] In the model training module, the inertia trend structure and the control strategy output are jointly used as feature inputs to determine whether the control strategy has a benchmark memory deviation.

[0045] Based on the identified behavior 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 from each tail trajectory marked as having an inertia trend as main feature parameters. Among them, the control channel set represents the channel number and identification actually participating in the trend convergence in the inertia trend, and all channels are no longer used, but only the key channels that produce trend convergence are used to establish training samples. The path convergence period is defined as the time length required for the control parameter change trend in the tail trajectory to stabilize and point to a certain trend center template. Its calculation method is as follows: from the start 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 confidence continuously rises and stabilizes near a certain center template (for example, the confidence is greater than 0.85 and the continuous time length is more than the set window length) in multiple consecutive time windows, it is determined that the convergence process is complete. The period length not only reflects the response speed of trend convergence, but also helps to judge the stable adjustment speed of the control system after disturbance. The extraction of the slope of the control parameter change curve is one of the key input features. The original curve of the control parameter change with time is constructed on each channel of the tail trajectory, and the first derivative sequence in the curvature change process is extracted through local window fitting. The slope sequence not only describes the parameter change speed, but also reflects the sensitivity and inertia of the system adjustment. Each channel outputs multiple quantitative features such as average slope, maximum slope, and slope coefficient of variation as part of the final input vector.

[0046] After the above feature extraction is completed, the system integrates the above content with the disturbance label information, disturbance factor trend and automatic control unit strategy output parameter in the corresponding disturbance response data set to generate training samples in a unified format. The training sample structure must ensure consistent dimensions and field standardization to avoid problems such as inconsistent dimensions or data drift during model training. The sample label setting is derived from the system's judgment result of whether the tail trajectory is in the "reference offset state" in the early stage. The sample identified as having reference offset is marked as a positive sample, and the trajectory without obvious trend convergence or strategy offset is marked as a negative sample to ensure that the model has distinguishing ability. The judgment of the reference offset state is based on the path offset direction. The direction of "positive offset" or "negative offset" is identified as a positive sample, and the direction of "undefined" is identified as a negative sample.

[0047] The structure of the trained model is a classification model based on a time series residual aggregation mechanism. This mechanism not only classifies the trend features at a single time, but also inputs the trend evolution feature sequence of each time window in the tail trajectory in sequence, and realizes the global judgment of the overall trajectory trend through the accumulation and aggregation of the time dimension residual. In the training process, the model learns the static boundaries between features, and can also capture the dynamic evolution law of the control behavior trend in the disturbance response process. It ensures the ability to identify time correlation and the accuracy of trend classification, and is suitable for the feature structure of "slow sliding of non-instantaneous mutation in path offset" in energy control behavior. In order to improve the model generalization ability, the system uses a multi-round cross-training mechanism in the training stage. All sample sets will be divided into multiple subsets, which will be rotated as validation sets and training sets. After each round of training, the system evaluates the accuracy, recall rate and offset tendency judgment confidence of the model, and selects the optimal parameter configuration in multiple rounds to freeze the final model. It ensures that the model is not a simple memory of the training set, but truly has the ability to identify trends and judge offset tendencies.

[0048] After the training is completed, the model will be deployed in a real-time recognition system. And for the tail track features generated during real-time operation, a benchmark memory offset judgment is performed. On the basis of the aforementioned completed data cleaning and channel structuring processing, it is ensured that the tail track contains a stable convergence trend and has sufficient control behavior information. Subsequently, the feature parameters of the tail track in all control channel dimensions are extracted, including but not limited to the slope information of the control parameter change curve, the channel activation sequence, the inter-channel collaborative change pattern, and the convergence period of the evolution path, etc. These features will be input into the trajectory trend recognition model as a unified vector structure. During the recognition execution process, the model maps the tail track vector to each offset path trend template space and calculates the fitting degree between it and each trend center template. This fitting degree is usually represented as a matching score or distance indicator, representing the convergence credibility of the tail track in the corresponding trend path. In order to provide structured offset judgment basis, the model outputs the fitting results of each path as normalized confidence labels. Each label corresponds to the matching credibility of a trend path, and the one with the highest confidence is the most likely behavior evolution category to which the current tail track belongs.

[0049] In the present embodiment, the automatic control unit refers to the key component in new energy equipment that undertakes actual control decision and execution function, which is usually deployed in the core control loop, analyzes the current state of the equipment in real time through preloaded control logic or strategy algorithm, and outputs parameter instructions for controlling the execution of each component. During the operation of 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 the embedded operation strategy, and generates new control decisions accordingly.

[0050] The control parameters output by the automatic control unit are strategy output parameters, which have the following characteristics: first, the strategy output parameters are not directly derived from human-computer interaction input or external adjustment commands, but are automatically calculated and generated by the control unit's internal preset control algorithm; second, such parameters are usually based on multi-channel input signals of equipment state, such as temperature, current, voltage, speed, load pressure, etc., and are calculated through embedded control models (such as PID control logic, fuzzy control logic or adaptive models based on state feedback), so they have high responsiveness and timeliness; finally, strategy output parameters not only include target set values, but also may include adjustment increments, execution channel selection signals, constraint boundary parameters, etc.

[0051] In the reset maintenance module, when the benchmark offset state is identified, an automatic reset command sequence is constructed according to the identification results, and parameter rollback instructions are issued to the control strategy.

[0052] The trajectory trend recognition model outputs a corresponding deviation path recognition result for the fitting result of the tail trajectory in the trend classification space. The result includes the belonging deviation path and a confidence label for quantifying the recognition confidence. The confidence label is generated based on the distance convergence degree, change direction consistency, curvature matching degree and other feature combinations of the trend vector of each channel of the tail trajectory in the trend space, and has trend deviation reflection capability. After obtaining the label, the deviation confidence of each channel is compared with the preset confidence threshold one by one. 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 sample statistics in the non-deviation state, so as to ensure that the deviation recognition does not occur false alarm or miss. When the confidence value of a certain channel exceeds the threshold, the current control state of the channel is marked as a parameter deviation state, and the subsequent automatic reset adjustment process is triggered. The judgment process must be executed channel by channel, and the cross-influence relationship between multiple channels should be considered to avoid local fluctuation interference to the overall recognition judgment.

[0053] Once it is determined that the deviation confidence of a certain channel exceeds the threshold, the current control state of the channel needs to be marked. The state marking should be included in the execution trigger condition of the subsequent parameter backtracking and reset command to ensure that the reset operation is not misused or missed. At the same time, the record information of the deviation state needs to be structured, including the marking time point, the corresponding disturbance segment identifier, the current control parameter value, the corresponding trend label and the fitting deviation direction, to provide complete background data support for subsequent reset path analysis and adjustment action execution. Backtrack the channel at all adjustment instruction sequences from the disturbance introduction point to the current time to extract the compensation path of the modified parameter.

[0054] After the parameter offset state is marked, a rollback operation is immediately performed to analyze the entire adjustment behavior process of the channel from the disturbance introduction point to the current time. The core of this process is to extract the correction path from the adjustment instruction sequence, i.e., the passive or active adjustment behavior set generated by the device in the process of attempting to maintain stability. In the operation, first, all adjustment control instructions in this time period need to be extracted, arranged in chronological order, and the parameter variation amplitude, feedback delay, and interaction between channels corresponding to each adjustment instruction need to be counted. Combined with the control target, it is determined whether it belongs to the attempt to correct 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 oscillation correction, sustained drift adjustment, and phase stability process. These change patterns are classified and labeled to facilitate the extraction of the reset reference point in the future. 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 feedback control response change direction corresponding to each adjustment instruction is counted after the instruction is issued, and it is determined whether the change shows a trend of returning to the stable interval; if the change direction of the adjusted control parameter is opposite to the tail trajectory offset direction, and the parameter values at multiple subsequent time points continuously approach the stable baseline before the disturbance introduction point, the instruction can be marked as an attempt to correct behavior. Further, by connecting the sequence of all instructions marked as correction behavior in chronological order, the correction compensation path of the channel can be constructed. This path reflects the tendency of the device to return to stability automatically initiated by the control system under the offset state.

[0055] After the compensation path is established, the most significant instantaneous point representing the device's deviation from the reference state is identified. This point usually represents the maximum change rate of the control parameter in unit time, i.e., the maximum slope point or inflection point. To identify this point, the parameter difference change amplitude before and after each time point can be calculated based on a sliding window, and the time point corresponding to the maximum change amplitude is selected as the maximum offset gradient point. Before this point, there is usually a stable interval of parameters, which can be determined by identifying a number of consecutive sampling points with parameter change rates below a set stability threshold. This stable parameter value is considered as the possible running reference state of the channel, denoted as the reset reference value. Next, a reset adjustment command sequence is constructed with the reference value as the target and the current control parameter difference, which gradually guides the parameter back to avoid rapid rollback and trigger control oscillation. After the construction is completed, the command sequence is issued to the channel control unit of the control system to force the execution of the parameter rollback and ensure that the control logic returns to the stable strategy framework. After the reset is executed, the system needs to continuously monitor the channel behavior to determine whether it has successfully returned to stability, and if not, manual intervention maintenance needs to be performed.

[0056] The above formulas are all de-dimensioned to calculate their numerical values. The formulas are obtained by collecting a large amount of data to simulate a formula of the most recent real situation. The preset parameters and threshold values in the formulas are set by a person skilled in the art according to actual conditions.

[0057] The above embodiments can be implemented wholly or partially by software, hardware, firmware, or any other combination. When implemented by software, the above embodiments can be implemented wholly or partially in the form of 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, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server, or data center to another by wired (for example, infrared, wireless, microwave, etc.) or wireless means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0058] Those skilled in the art can realize that the modules and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. A person 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 the present application.

[0059] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described system, device, and module can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0060] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the division of the above-described device embodiment is merely an example, and there can be other division manners. For example, the modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or modules, and can be in electrical, mechanical or other forms.

[0061] The modules illustrated as separated components can or can not be physically separated, and the components illustrated as modules can or can not be physical modules, and can be located in one place, or can be distributed on a plurality of network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment.

[0062] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can be physically present alone, or two or more modules can be integrated into one module.

[0063] If the functions are realized in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program codes that can be stored in the medium.

[0064] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0065] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included in the scope of protection of the present application.

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 instruction sequence of the new energy equipment during operation and construct a two-dimensional behavior trajectory matrix reflecting the timing of the control actions; The disturbance mapping module is used to collect disturbance factor data and establish a mapping relationship between disturbance factors and behavior trajectories; The deviation identification module is used to identify whether the tail segment trajectory deviates from the initial trajectory based on the tail segment trajectory state after the disturbance phase, and record the deviation path of the corresponding control channel; The inertia recognition module is used to extract the multi-channel evolution trend vector of the tail trajectory when the trajectory feature offset is identified, and to construct a channel-level inertia trend structure; The model training module is used to train the trajectory trend recognition model, combining the inertial trend structure and the control strategy output into a feature input to determine whether the control strategy has a baseline memory offset; The reset maintenance module is used to construct an automatic reset command sequence based on the back-adjustment path according to the identification result when it is identified as a reference offset state, 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 is characterized in that: In the trajectory extraction module, the control instruction sequence during the operation of the new energy equipment is collected and a two-dimensional behavior trajectory matrix reflecting the timing of the control action is constructed, which specifically includes: Collect all control instructions of new energy equipment during the disturbance response cycle, and extract the control parameters, control channel identifier, feedback response delay and corresponding execution timestamp in each instruction; Arrange the extracted instruction sequence in chronological order and divide the channel groups according to the control channel; Within each channel group, the control action segments are divided based on a sliding window with a fixed time width. The control parameter change amplitude, control triggering times, and average feedback response delay within each segment are counted and integrated into a channel-level control behavior vector. The behavior vectors of each channel are spliced ​​in time sequence 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 is characterized in that: In the disturbance mapping module, collecting disturbance factor data and establishing a mapping relationship between the disturbance factor and the behavior trajectory specifically includes: Extracting disturbance factor data associated with the new energy equipment from the Internet of Things sensing network, wherein the disturbance factor corresponds to a functional type of the new energy equipment; Based on the first-order derivative of the change rate of the disturbance factor data and the disturbance amplitude threshold, the disturbance introduction point and disturbance relief point are identified, and the stage labels of the continuous disturbance are divided into the introduction period, stable period and decay period according to the duration and change form of the continuous disturbance; Map the divided disturbance stage labels to the time axis of the behavior trajectory matrix in a time-aligned manner; Record the parameter changes, control channel trigger frequency, and local statistical characteristics of feedback response delay of the behavioral trajectory segment corresponding to each disturbance stage; The disturbance stage labels, the change rate of disturbance factor data and the statistical characteristics of the behavior trajectory fragments 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 is characterized in that: In the deviation identification module, based on the state of the tail segment trajectory after the disturbance phase, identifying whether the tail segment trajectory deviates from the initial trajectory and recording the deviation path of the corresponding control channel specifically include: A fixed-length behavioral trajectory segment after the end of the disturbance decay period is intercepted from the behavioral trajectory matrix as the tail segment trajectory; Extract the control parameters of each channel of the tail trajectory, calculate the activation order of each channel, the amplitude of the control parameter change and the consistency of the feedback response delay, and construct the state vector of the tail trajectory; The initial behavior trajectory segment with the same length as the tail trajectory before the disturbance introduction point is called, and the initial behavior trajectory state vector is extracted using trajectory compression alignment, and the multi-dimensional channel vector difference calculation is performed with the tail trajectory state vector; According to the preset stability index interval, the mean and variance stability test is 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 deviation; 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 is characterized in that: In the inertia recognition module, when a trajectory feature offset is identified, the multi-channel evolution trend vector of the tail trajectory is extracted, and the channel-level inertia trend structure is constructed, which specifically includes: Extract the time evolution curve of each channel control parameter output from the offset path marked as the trajectory feature offset state; In the control parameter change curve of each channel, an evolution trend sequence is constructed based on the change in the curve slope, and the channel evolution trends under different disturbances are grouped according to the disturbance stage label; A clustering method is used to establish a path change pattern classification set for the trend vector sequence. The curvature similarity of the control parameter curves corresponding to the channels in the group is divided and the trend center template is extracted to construct the 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 simultaneously in multiple channels, 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 is 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 combined to form a feature input. The determination of whether the control strategy has a baseline memory offset specifically includes: The channel, path convergence period and control parameter change curve slope extracted from the behavioral inertia trend structure are used as feature inputs; Combine the disturbance response dataset with the policy output parameters of the automatic control unit to construct training samples in a unified format; A classification model construction method based on the time series residual aggregation mechanism is adopted to conduct multiple rounds of cross-training on the training samples to establish a trajectory trend recognition model; The tail segment trajectory features to be identified are input into the trajectory trend recognition model, the fitting degree of the output results in each offset path is calculated, and the confidence label of the potential baseline memory offset is output.

7. The intelligent operation and maintenance management system for new energy equipment based on the Internet of Things according to claim 6 is characterized in that: The automatic control unit refers to a component in the new energy equipment that automatically performs decision-making control, and 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 is characterized in that: In the reset maintenance module, when the reference offset state is identified, the automatic reset command sequence is constructed based on the backtracking adjustment path according to the identification result, and the 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 a channel where the confidence exceeds a preset offset threshold, the current control state of the channel is marked as a parameter offset state; The channel traces back all the adjustment instruction sequences from the disturbance introduction point to the current moment and extracts the compensation path of the correction parameters; The maximum offset gradient point of the correction parameter is identified in the compensation path, and the parameter of the previous stable segment at this point is used as the reset reference value to construct an automatic reset adjustment command sequence, and the reset instruction execution strategy output parameter rollback is issued to the controlled channel.

Images