A setting machine flue gas waste heat recovery system and method

Abstract

The present application relates to setting machine waste heat recovery technical field, disclose a kind of based on setting machine flue gas waste heat recovery system and method.The method is continuously obtained by the temperature, pressure, flow reading of flue gas by the sensor array installed on setting machine flue gas pipeline, while recording setting machine real-time power output and ambient humidity data, form current working condition feature set.Utilize the feature set in historical database search similar feature historical operation record, typical working condition category is divided by mode identification processing, extract and analyze corresponding heat recovery efficiency data sequence and operating parameter sequence, calculate performance evaluation value sequence.It is input time series analysis module, and output heat recovery trend index and heat energy loss trend index.By comparison with typical working condition category benchmark feature, dynamically correct index, obtain calibrated heat recovery rate and calibrated heat energy loss rate, generate control parameter according to this, adjust waste heat recovery equipment operating state.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of setting machine waste heat recovery, in particular to a setting machine flue gas waste heat recovery system and method. BACKGROUND

[0002] In the production process of textile printing and dyeing, the energy consumption of the setting machine accounts for a large proportion of the total energy consumption. These energy consumptions are mainly used to maintain the high temperature environment inside the setting machine, and a large amount of heat energy is directly discharged with flue gas, causing serious waste. In order to alleviate energy waste, most industries use two types of traditional waste heat recovery equipment, namely gas-gas heat exchange and gas-water heat exchange, but both have obvious defects. The gas-gas heat exchange equipment is large in size, difficult to install, high in wind resistance during operation, and prone to heat exchange efficiency decline and maintenance cost increase due to high temperature and dust. The gas-water heat exchange equipment is prone to scaling, complex installation and maintenance, and due to the thermal inertia of water, it cannot adapt to the change of setting machine working conditions in time and has slow response.

[0003] In addition, some industries try to introduce an intelligent control waste heat recovery system, which only uses fixed set values or simple PID control, which is difficult to cope with the complex working conditions of the setting machine. The setting machine has strong nonlinearity and time-varying nature, and factors such as flue gas temperature and flow interact to cause the working point to drift frequently. Fixed strategies cannot maintain optimal efficiency in all working conditions, and simple feedback control has hysteresis, which cannot anticipate the trend of heat recovery and heat loss, and cannot dynamically correct the index according to the difference in working conditions, resulting in poor working condition adaptability and slow dynamic response, making it difficult to continuously and efficiently recover waste heat.

[0004] If the pollutants such as particulate matter and volatile organic compounds contained in the flue gas are not treated before being discharged, environmental risks will be caused. However, the existing technology has always failed to solve the core problem - it cannot output accurate trend indicators through time series analysis, cannot dynamically correct the index based on the difference between the current and historical typical working conditions to obtain the calibration rate, and it is even more difficult to generate control parameter adjustment equipment according to the calibration rate, resulting in unstable waste heat recovery efficiency, which cannot meet the energy saving and compliance needs of enterprises. SUMMARY

[0005] The present application aims to provide a setting machine flue gas waste heat recovery system and method to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides a setting machine flue gas waste heat recovery method, which comprises:

[0007] The temperature readings, pressure readings and flow readings of the flue gas are continuously obtained by the sensor array installed on the flue gas pipeline of the setting machine, and the real-time power output and ambient humidity data of the setting machine are recorded to form a current working condition feature set;

[0008] searching for historical operation records with similar features in the historical database using the current working condition feature set, and performing pattern recognition processing on the historical operation records to divide them into multiple typical working condition categories; for each typical working condition category, extracting the corresponding historical heat recovery efficiency data sequence and operation parameter sequence, and calculating the performance evaluation value sequence under each category through statistical analysis;

[0009] inputting the performance evaluation value sequence into a time series analysis module to output a heat recovery trend indicator and a heat energy loss trend indicator; calculating the similarity between the current working condition feature set and the reference features of each typical working condition category, selecting the category with the highest similarity as the reference category, and dynamically correcting the heat recovery trend indicator and the heat energy loss trend indicator according to the difference between the current working condition feature set and the reference features of the reference category to obtain the calibrated heat recovery rate and the calibrated heat energy loss rate;

[0010] generating control parameters according to the calibrated heat recovery rate and the calibrated heat energy loss rate to adjust the operating state of the waste heat recovery equipment.

[0011] Preferably, the temperature readings, pressure readings and flow readings of the flue gas are continuously obtained by a sensor array, while the real-time power output of the forming machine and the ambient humidity data are recorded to form the current working condition feature set, specifically including:

[0012] The instantaneous temperature value at the flue gas inlet is measured using a temperature sensor, the real-time pressure value in the flue gas pipeline is measured using a pressure sensor, the flue gas volume flow value is monitored using a flowmeter, the power output percentage is read from the forming machine controller, and the humidity percentage value is obtained from an environmental sensor; the temperature value, pressure value, flow value, power output percentage and humidity percentage value are aligned by timestamp and combined into a multi-dimensional vector to constitute the current working condition feature set.

[0013] Preferably, the current working condition feature set is used to search for historical operation records with similar features in the historical database, and pattern recognition processing is performed on these historical operation records to divide them into multiple typical working condition categories, specifically including:

[0014] The Euclidean distance between the current working condition feature set and the feature vector of each historical operation record in the historical database is calculated, and the historical operation records with a distance less than a threshold value are selected as an initial similar set; all historical operation records in the initial similar set are grouped using a clustering algorithm, and the clustering is based on the combined features of temperature readings, pressure readings and power output values to form multiple typical working condition categories.

[0015] Preferably, for each typical working condition category, the corresponding historical heat recovery efficiency data sequence and operation parameter sequence are extracted, and the performance evaluation value sequence under each category is calculated through statistical analysis, specifically including:

[0016] retrieve historical heat recovery efficiency values from each member record of the typical working condition category, and arrange them in chronological order to form a historical heat recovery efficiency data sequence; meanwhile, extract the operating parameter values at the corresponding time, including the fan speed and the valve opening, to form an operating parameter sequence; perform weighted average calculation on the historical heat recovery efficiency data sequence and the operating parameter sequence, with the weights determined based on the deviation of the operating parameters from the standard values, and output a performance evaluation value sequence.

[0017] Preferably, the weighted average calculation on the historical heat recovery efficiency data sequence and the operating parameter sequence, with the weights determined based on the deviation of the operating parameters from the standard values, outputs a performance evaluation value sequence, specifically including:

[0018] Based on the deviation of the operating parameter sequence from the preset standard value range, determine the dynamic weight coefficient corresponding to each operating parameter through an adaptive weight calculation mechanism;

[0019] Multiply each efficiency value in the historical heat recovery efficiency data sequence by the dynamic weight coefficient at the corresponding time point to obtain a weighted efficiency value sequence;

[0020] Perform denoising processing on the weighted efficiency value sequence to eliminate interference components, and then perform smoothing estimation to optimize data stability, finally output a performance evaluation value sequence.

[0021] Preferably, the time series analysis module is configured to perform:

[0022] Perform difference operation on the performance evaluation value sequence to calculate the change rate at adjacent time points, and obtain a change rate sequence; extract the rising section and the falling section from the change rate sequence, and take the average change rate of the rising section as the heat recovery trend indicator and the average change rate of the falling section as the heat energy loss trend indicator;

[0023] The difference operation adopts a first-order backward difference formula, and the change rate sequence is subjected to moving average filtering to eliminate noise.

[0024] Preferably, the difference between the current working condition feature set and the reference category benchmark feature is used to dynamically correct the heat recovery trend indicator and the heat energy loss trend indicator, to obtain a calibrated heat recovery rate and a calibrated heat energy loss rate, specifically including:

[0025] Construct a geometric space model of the feature vector, and map the current working condition feature set and the benchmark features of each category to points on the manifold;

[0026] Calculate the spatial distance between the current working condition feature point and the benchmark feature points of each category, and select the category with the smallest distance as the reference category;

[0027] Establish a linear approximation space corresponding to the current working condition feature point, project the reference category benchmark feature point to the linear approximation space to obtain a projection vector;

[0028] component decomposition is performed on the projection vector to obtain an intensity component representing the intensity difference and a mode component representing the mode difference;

[0029] An intensity correction factor is determined based on the intensity component, and a mode correction parameter is determined based on the mode component;

[0030] The heat recovery trend index is corrected by combining the intensity correction factor and the mode correction parameter to obtain a calibrated heat recovery rate;

[0031] The heat energy loss trend index is corrected by combining the reverse adaptation parameter corresponding to the intensity correction factor and the mode correction parameter to obtain a calibrated heat energy loss rate.

[0032] Preferably, the control parameter is generated according to the calibrated heat recovery rate and the calibrated heat energy loss rate, and the operating state of the waste heat recovery device is adjusted, specifically including: mapping the calibrated heat recovery rate to a fan speed set value, and mapping the calibrated heat energy loss rate to a valve opening set value; generating a control signal containing the fan speed set value and the valve opening set value, and sending it to the actuator of the waste heat recovery device.

[0033] Preferably, all historical operating records in the initial similarity set are grouped using a clustering algorithm, and the clustering is based on the combined characteristics of temperature readings, pressure readings and power output values, wherein the clustering algorithm uses a mean shift method to automatically determine the number of categories.

[0034] Preferably, the present application also includes a setting machine flue gas waste heat recovery system, which comprises a memory, a processor and a computer program stored in the memory and running on the processor, and the processor implements the steps of the setting machine flue gas waste heat recovery method when executing the computer program.

[0035] Compared with the prior art, the present application has the following advantages:

[0036] The performance evaluation value sequence is input into the time series analysis module, and the heat recovery trend index and the heat energy loss trend index are output, which are like "early warning" and "compass" of the waste heat recovery system. The operator can know the development trend of heat recovery and the possible heat energy loss in advance, so as to take timely measures for optimization. By calculating the similarity between the current working condition feature set and the reference feature of each typical working condition category, the category with the highest similarity is selected as the reference category, and the heat recovery trend index and the heat energy loss trend index are dynamically corrected according to the difference between the current working condition feature set and the reference category reference feature, to obtain the calibrated heat recovery rate and the calibrated heat energy loss rate. This dynamic correction method can closely match the real-time changing operation condition of the forming machine, ensure the accuracy of the heat recovery rate and the heat energy loss rate, further optimize the operation of the waste heat recovery equipment, and maximize the reduction of heat energy loss and improve the energy recycling rate.

[0037] The control parameters are generated according to the calibrated heat recovery rate and the calibrated heat energy loss rate, and the operation state of the waste heat recovery equipment is adjusted. The waste heat recovery equipment can adjust its operation state in real time according to these accurately generated control parameters, realizing efficient use of energy. Enterprises no longer need to spend a lot of energy cost on waste heat recovery, but only get low recovery efficiency, but can efficiently recover waste heat while reducing energy cost and improving economic efficiency of enterprises. The waste heat recovery equipment reduces the emission of waste heat in the process of efficient operation. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 The working principle diagram of the waste heat recovery method based on the forming machine is described in the present application.

[0039] Figure 2 The flow chart of the specific steps of the mean shift clustering grouping is described.

[0040] Figure 3 The flow chart of the specific steps of calculating the performance evaluation value sequence is described.

[0041] Figure 4 The heat recovery efficiency and operation parameter relationship diagram is described.

[0042] Figure 5 The heat recovery trend index and the heat energy loss trend index change with the running time trend diagram. DETAILED DESCRIPTION

[0043] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described, obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0044] Please refer to Figure 1 The present application provides a setting machine flue gas waste heat recovery system and method, the method comprises: through the sensor array installed on the setting machine flue gas pipeline, continuously obtaining the temperature reading, pressure reading and flow reading of the flue gas, at the same time, recording the real-time power output of the setting machine and the ambient environment humidity data, these data are integrated as the current working condition feature set, the feature set is stored in the form of multi-dimensional vector, and is updated with time stamp. Using the current working condition feature set to search the historical operation record with similar features in the historical database, the historical database contains a large amount of past operation data, the search process is based on feature similarity, then the matched historical record is subjected to pattern recognition processing, a plurality of typical working condition categories are divided by clustering algorithm, each category represents a common operating state. For each typical working condition category, extract its corresponding historical heat recovery efficiency data sequence and operation parameter sequence, the operation parameters include fan speed and valve opening, etc., calculate the performance evaluation value sequence under each category through statistical analysis, the sequence reflects the heat recovery performance in the historical operation. Then input the performance evaluation value sequence into the time series analysis module, the module performs difference operation and trend extraction, outputs the heat recovery trend index and the heat energy loss trend index, these indexes represent the change direction of heat recovery efficiency. At the same time, the current working condition feature set is calculated with the reference feature of each typical working condition category, the reference feature is the representative feature vector of each category, the category with the highest similarity is selected as the reference category through distance measurement, and the heat recovery trend index and the heat energy loss trend index are dynamically corrected according to the difference between the current working condition feature set and the reference category reference feature, the correction process involves intensity and mode adjustment, to obtain the calibrated heat recovery rate and the calibrated heat energy loss rate. According to the calibrated heat recovery rate and the calibrated heat energy loss rate, control parameters are generated, these parameters are mapped as the set values of the waste heat recovery equipment, such as fan speed and valve opening, the equipment operating state is adjusted through the control signal, to realize real-time optimization.

[0045] In specific implementations, the temperature sensor employs a K-type thermocouple with a measurement range covering zero Celsius to eight hundred Celsius, which can meet the monitoring requirements of the setting machine flue gas temperature. The temperature sensor is fixed to the flue gas pipe wall through threaded connection or flange installation. The installation position is selected at the center area of the representative cross section to avoid measurement errors caused by pipe wall effect. The temperature sensor signal output is an analog signal, such as a four-milliamper to twenty-milliamper current signal or a zero-volt to five-volt voltage signal. The signal is transmitted to the data acquisition module through a shielded cable. In some embodiments, the temperature sensor employs a non-contact infrared temperature detector. The infrared temperature detector detects the infrared radiation intensity of the flue gas in real time through a window and converts it into a temperature reading. This way avoids the wear and tear caused by direct contact with high-temperature flue gas. The data acquisition module periodically samples the temperature signal at a sampling frequency of one hundred hertz. Each sampling point obtains an instantaneous temperature value. The instantaneous temperature value is converted to digital by an analog-to-digital converter and is marked with a high-precision time stamp. The time stamp is provided by the system clock with a precision of milliseconds.

[0046] The pressure sensor measures the real-time pressure value in the flue gas pipe. In specific implementations, the pressure sensor selects a piezoresistive pressure transmitter with a range of zero Pascal to five thousand Pascal. The pressure sensing port of the pressure sensor is connected to the inside of the flue gas pipe through a pressure guide pipe. The pressure sensing port is perpendicular to the gas flow direction to accurately sense the flue gas static pressure. The pressure sensor outputs a standard analog signal. The signal is amplified and filtered by a signal conditioning circuit to eliminate high-frequency interference. The data acquisition module synchronously acquires the pressure signal with the same sampling frequency as the temperature signal acquisition. Each pressure reading has the same time stamp as the corresponding temperature reading, ensuring data time sequence consistency. It can be understood that multiple pressure sensors can be deployed at different positions of the flue gas pipe, such as at the pipe inlet, outlet, and possible pressure loss bends, to obtain the internal pressure distribution of the pipe. The readings of multiple pressure sensors can calculate the average pressure value or monitor the pressure gradient change.

[0047] The flow meter monitors the flue gas volume flow value. In specific embodiments, the flow meter is a differential pressure flow meter or a vortex flow meter. The differential pressure flow meter generates a pressure difference through a throttling device and calculates the volume flow according to Bernoulli's equation. The vortex flow meter determines the flow size by detecting the vortex frequency generated by the fluid passing through the bluff body. The flow meter is installed in the straight pipe section of the flue gas pipeline. The installation position meets the requirement of the straight pipe section of the first ten times the pipe diameter and the last five times the pipe diameter to ensure the accuracy of the flow measurement. The flow meter outputs a pulse signal or an analog signal. The pulse signal frequency is proportional to the flow. The analog signal is a four to twenty milliampere current signal representing the flow value in the zero to maximum range. The data acquisition module counts or converts the flow signal to obtain the flue gas volume flow value. The volume flow value is in cubic meters per hour, and the reading is updated every second. It can be understood that in the high-temperature flue gas environment, the flow meter needs to have high-temperature resistance, such as using high-temperature alloy material or installing a heat dissipation device to ensure long-term stable operation.

[0048] The power output percentage is read from the former controller. In specific embodiments, the former controller uses a programmable logic controller or a distributed control system. The power output percentage reflects the load rate of the main motor of the former. It is calculated by monitoring the motor current or the frequency output of the frequency converter. The power output percentage value is transmitted through an industrial communication network. The communication protocol uses Modbus TCP / IP or PROFIBUS DP. The communication cycle is set to one hundred milliseconds. The data acquisition module acts as the master station to initiate the read request and obtains the power data from the slave station of the former controller. The power output percentage value is between zero and one hundred. It represents the ratio of the current load to the rated load. The value is updated every two hundred milliseconds. The update time is synchronized with the sensor data acquisition time through the network time protocol. In some embodiments, the power output percentage can be obtained from the former upper monitoring system through the OPC protocol. The upper system displays and stores the device operating parameters in real time. The data acquisition module subscribes to the relevant data points through the OPC client interface.

[0049] The humidity percentage value is obtained from the environmental sensor, which in a specific implementation is a capacitive humidity sensor installed in a well-ventilated area within five meters of the equipment, avoiding the influence of local heat sources or air flow dead zones. The humidity sensor measures the relative humidity of the air, with a range of zero to one hundred percent relative humidity, and outputs an analog signal of four to twenty milliamps corresponding to zero to one hundred percent humidity values. The data acquisition module collects humidity signals at one-second intervals, and after analog-to-digital conversion and linearization processing, obtains the humidity percentage value, which represents the ratio of the water vapor partial pressure in the air to the saturated water vapor partial pressure. Optionally, the environmental sensor can integrate temperature and atmospheric pressure measurement functions to provide more comprehensive environmental parameter information. The integrated sensor outputs data packets through a digital interface such as I2C or RS-485, and the data packets contain humidity, temperature, and multi-parameter measurement values. The temperature value, pressure value, flow value, power output percentage, and humidity percentage value are aligned by timestamp. In a specific implementation, the data acquisition module has a data buffer area inside, which can store the last three hundred records, each record containing a timestamp and corresponding multi-parameter measurement values. The timestamp is generated by a high-precision real-time clock chip, which is continuously running through battery backup. The timestamp format is year, month, day, hour, minute, second, and millisecond. After all sensor data arrives at the acquisition module, it is inserted into the corresponding position of the buffer area according to its timestamp. The alignment algorithm takes the earliest arriving data time as the reference to find other parameter data within the same time window. The time window size is set to ten milliseconds, and the data falling within the same window is considered as synchronous data. The aligned data is combined into a multi-dimensional vector, with the vector form being [temperature value, pressure value, flow value, power output percentage, humidity percentage]. The vector dimension is fixed at five, and each dimension corresponds to a measurement parameter. Optionally, the data alignment process can be completed in the upper computer. Each sensor data is transmitted to the upper computer through Ethernet, and the upper computer runs data integration software. The software performs interpolation alignment according to the timestamp in the data packet to generate a synchronous data vector.

[0050] The merged multidimensional vectors constitute the current operating condition feature set. In practice, this feature set is stored in random access memory using a circular queue data structure. The queue length is set to the amount of data from the most recent hour, i.e., 3,600 records, with each record being a five-dimensional vector. The current operating condition feature set is updated over time, with new data inserted from the tail of the queue and the oldest data removed from the head when the queue is full, ensuring the dataset always reflects the latest operating status. The current operating condition feature set supports fast querying and batch access, providing a data source for subsequent similarity searches and pattern recognition. The feature set data is periodically backed up to non-volatile memory, with a backup cycle that can be set to one hour or twenty-four hours to prevent data loss due to unexpected power outages. It is understood that the current operating condition feature set can be expanded to include more parameters, such as flue gas component concentration or equipment vibration data, simply by adding vector dimensions and corresponding sensor channels; the system architecture has good scalability.

[0051] Example 2: See Figure 2 The system searches a historical database for historical operation records with similar characteristics using the current operating condition feature set. In practice, the historical database is constructed using a time-series database architecture for efficient storage and querying of time-series data. The historical database runs on a dedicated server and is connected to the data acquisition system via gigabit Ethernet. Each historical operation record in the database contains a complete timestamp field and multiple numerical fields. The timestamp field records the exact time the data was generated, and the numerical fields include historical temperature readings, historical pressure readings, historical flow rates, historical power output percentages, and historical ambient humidity values. Each historical operation record also includes a unique record identifier. The historical database tables are partitioned according to time ranges, such as by month or year. Each partition has an independent index, with index fields including timestamps and key parameter values ​​to accelerate range and conditional queries. The database query interface provides standardized SQL or SQL-like query language support, allowing client programs to submit complex query conditions. It is understood that the historical database requires regular maintenance, including data backup, index rebuilding, and defragmentation, to ensure query performance and data security.

[0052] The Euclidean distance between the current working condition feature set and the feature vector of each historical running record in the historical database is calculated. In specific embodiments, the current working condition feature set is a multi-dimensional vector containing five elements, which correspond to the real-time temperature value, real-time pressure value, real-time flow value, real-time power output percentage, and real-time humidity percentage value, respectively. The feature vector of the historical running record is a multi-dimensional vector containing the same five parameter values recorded at a certain time in history. The Euclidean distance between the two vectors is calculated by a specific formula involving the square sum of the value difference in each corresponding dimension and then taking the square root. The calculation process is performed by a dedicated distance calculation module, which reads the current working condition feature set vector from memory, and then iterates through each historical running record in the historical database to calculate the Euclidean distance with the current vector one by one. The distance calculation module uses a batch processing method, loading a batch of historical records from the historical database into the memory buffer each time, and calculating a batch before loading the next batch to reduce the number of database accesses. In some embodiments, in order to process large-scale historical data, the distance calculation can use a distributed computing framework to shard the historical data to multiple computing nodes for parallel Euclidean distance calculation, and finally aggregate the calculation results of all nodes.

[0053] A dynamic threshold is set and historical running records with a distance less than the threshold are selected as the initial similar set. In specific embodiments, the dynamic threshold is automatically adjusted according to the distribution characteristics of the historical data. The threshold calculation module first calculates the mean and standard deviation of the Euclidean distances between all feature vectors in the historical database, and then determines the threshold size based on the mean plus several times the standard deviation. The threshold calculation module recalculates the dynamic threshold periodically, with a recalculating period of one day or one week to adapt to the slow changes in the distribution of historical data. The selection process is performed by the record filtering module, which iterates through all calculated Euclidean distance values and adds the historical running record identifiers with a distance value less than the dynamic threshold to the candidate list. The initial similar set is stored in memory in the form of a data set, each element in the set containing the unique identifier of the historical running record and the Euclidean distance value of the record with the current working condition feature set. Optionally, the dynamic threshold can be fine-tuned according to the characteristics of the current working condition feature set itself, for example, when the parameter fluctuations of the current working condition feature set are large, the threshold is appropriately increased to include more historical records and enhance system adaptability.

[0054] The clustering algorithm is applied to group all the historical run records in the initial similarity set. In some embodiments, the clustering algorithm uses a mean shift method, which is a kernel density estimation based unsupervised clustering algorithm that can automatically determine the number of natural classes in the data set without pre-specifying the number of clusters. The mean shift method processing starts from each data point in the initial similarity set, each data point represents a feature vector of a historical run record, which contains the combined features of temperature readings, pressure readings, and power output values. The mean shift algorithm defines a kernel function for each data point, typically using a Gaussian kernel function with a bandwidth parameter to control the influence range, which is automatically set according to the data density. The mean shift iteration process starts from each data point, calculates the density gradient of the data within a certain range around the point, then moves the data point along the gradient direction, the moving step is determined by the kernel function and the bandwidth, and the process is repeated until it converges to the point of maximum density. All original data points that converge to the same point of maximum density are grouped into the same cluster, and each cluster represents a typical operating condition category. It can be understood that the mean shift method has a certain robustness to noise points and isolated points in the initial similarity set, which will be ignored or grouped into edge categories during the iteration process, ensuring the stability of the clustering result.

[0055] The mean shift method automatically determines the number of categories, which is not a pre-set fixed value in some embodiments, but is generated as a natural result of the mean shift process. The number of converged points of maximum density is the final determined number of categories. Each point of maximum density becomes the cluster center of the corresponding category, and the feature vector of the cluster center is used as the reference feature of the category for subsequent similarity comparison. The clustering result evaluation module will check the number of members in each category, and the category with too few members may be considered as an abnormal category and be merged or removed, and the member number threshold is usually set to a certain percentage of the total data points. After the clustering algorithm is executed, the system assigns a unique category identifier to each typical operating condition category and establishes a mapping relationship between the category and the member historical run records, which is stored in the classification lookup table. In some embodiments, the mean shift method can be configured with different kernel function types and bandwidth parameters to adapt to different types of data distribution characteristics, such as adaptive bandwidth strategy for unevenly distributed data. Optionally, the clustering process can be performed periodically offline, such as re-clustering all historical data at low system load in the early morning every day, updating the definition of typical operating condition categories to reflect the long-term changes in device operation mode.

[0056] A plurality of typical operating condition categories are formed, each of which has a clear mathematical definition and a set of members in specific implementation, the category definition includes a cluster center vector and a category radius, which is the maximum Euclidean distance from the cluster center to all member points of the category. The typical operating condition category information is stored persistently in the category knowledge base, which uses a relational database table structure for storage, containing fields such as category identifier, cluster center vector, category radius, creation time, etc. When the system is running, the new current operating condition feature set can be quickly matched with these predefined typical operating condition categories, avoiding the need to perform time-consuming clustering calculations every time. Typical operating condition categories are incrementally updated according to newly added historical operation records, incremental update strategies include periodically assigning new historical records to the nearest existing category, or creating a new category when the distance between the new record and all existing category centers exceeds the threshold.

[0057] Embodiment 3: refer to Figure 3 For each typical operating condition category, the corresponding historical heat recovery efficiency data sequence and operating parameter sequence are extracted, in specific implementation, the extraction of the historical heat recovery efficiency data sequence is completed through a database query operation, the system sends a structured query statement to the historical database, the statement contains the identifier of the typical operating condition category as a filter condition, and the query result returns the heat recovery efficiency values of all historical operation records belonging to the category. The calculation of heat recovery efficiency value is based on the principle of energy balance, which is determined by the enthalpy difference between the inlet and outlet of flue gas and the input energy, each efficiency value is associated with an accurate timestamp. The historical heat recovery efficiency data sequence is arranged in chronological order according to the timestamp, forming a strictly monotonically increasing time sequence, each data point in the sequence contains two fields of timestamp and efficiency value. The extraction of the operating parameter sequence is carried out synchronously, the operating parameter sequence includes the fan speed value and the valve opening value, the fan speed value is obtained from the frequency converter operation log, the unit is revolutions per minute, the valve opening value is extracted from the actuator feedback record, which is expressed as percentage opening. The operating parameter sequence and the historical heat recovery efficiency data sequence are accurately matched by timestamp, ensuring that the efficiency value and the operating parameter value at the same time are correctly matched, the matched data is organized into a tuple sequence with time marker and stored in the memory array. It can be understood that the sequence length depends on the number of historical records contained in the typical operating condition category, the sequence length of different categories may be different, the system will maintain an independent data sequence set for each category.

[0058] The performance evaluation value sequence under each category is calculated through statistical analysis. In a specific implementation, the core of the statistical analysis is weighted average calculation, which is performed for each efficiency value in the historical heat recovery efficiency data sequence. The weight of each efficiency value is determined based on the deviation of the operating parameter value at the corresponding time from the preset standard value range. The preset standard value range is derived from the technical specifications of the equipment manufacturer or the optimal operating interval obtained from long-term operation statistics. For example, the standard value range of the fan speed may be 1,200 revolutions per minute to 1,500 revolutions per minute, and the standard value range of the valve opening may be 40% to 60%. The deviation calculation module first calculates the absolute difference between each operating parameter value and the median of the standard value range, and then combines the deviations of multiple operating parameters into a comprehensive deviation index. The weighted average calculation uses the following mathematical relationship:

[0059]

[0060] where ω i represents the dynamic weight coefficient of the i-th efficiency value, θ ij represents the actual value of the j-th operating parameter at the i-th time, μ j represents the median of the standard value range of the j-th operating parameter, σ j represents half of the width of the standard value range of the j-th operating parameter, and m represents the total number of operating parameters.

[0061] An adaptive fuzzy weight calculation module is constructed. In a specific implementation, the adaptive fuzzy weight calculation module takes the deviation of the operating parameter sequence from the preset standard value range as input, and the deviation value is converted into a fuzzy set through fuzzification processing. The fuzzification processing first defines the fuzzy set of the input variable. For example, for the fan speed deviation, five language values of "negative large", "negative small", "zero", "positive small", and "positive large" can be defined, each of which corresponds to a membership function. The membership function adopts a triangular or trapezoidal function, and the function parameters are manually set according to the characteristics and importance of the operating parameters or optimized through machine learning. The fuzzy rule base includes a series of if-then rules. The antecedent of the rule is the fuzzy proposition of the operating parameter deviation, and the consequent of the rule is the fuzzy proposition of the weight adjustment. For example, "if the fan speed deviation is negative small and the valve opening deviation is zero, then the weight is high". The defuzzification process calculates the clear dynamic weight coefficient output using the barycenter method, and the dynamic weight coefficient is used as the input of the weighted average calculation. The parameters of the adaptive fuzzy weight calculation module can be adjusted online. When the system operating conditions change, the fuzzy rule base is updated or the membership function parameters are adjusted to adapt to the new operating environment. It can be understood that the adaptive fuzzy weight calculation module can handle the coupling relationship and nonlinear influence between operating parameters, and is more accurate in reflecting the actual influence of operating parameters on heat recovery efficiency than the simple linear weighting method.

[0062] The preset fuzzy rule base in the adaptive fuzzy weight calculation module includes the following typical rules (taking fan speed deviation ΔRPM and valve opening deviation ΔValve as examples):

[0063] Rule 1: If ΔRPM is 'negative large' and ΔValve is 'negative large', the weight coefficient is 'extremely high'.

[0064] Rule 2: If ΔRPM is 'zero' and ΔValve is 'positive small', the weight coefficient is 'high'.

[0065] Rule 3: If ΔRPM is 'positive large' or ΔValve is 'positive large', the weight coefficient is 'low'.

[0066] Wherein, the domain of ΔRPM can be set as [-200, 200] rpm, and the membership function of the fuzzy subset {negative large, negative small, zero, positive small, positive large} adopts a triangular function; the domain of the weight coefficient is [0.5, 1.5].

[0067] Each efficiency value in the historical heat recovery efficiency data sequence is multiplied by the dynamic weight coefficient at the corresponding time point to obtain a weighted efficiency value sequence. In specific implementation, the multiplication operation is performed by a scalar multiplier point by point, each historical heat recovery efficiency value is multiplied by the normalized dynamic weight coefficient to generate a weighted efficiency data point. The weighted efficiency value sequence maintains the time sequence and structure of the original sequence, but the numerical size of each point reflects the importance degree under the corresponding operating parameter condition. The weighted efficiency value sequence is temporarily stored in the cache area, waiting for subsequent signal processing operations. Abnormal values or obviously erroneous data points in the sequence will be detected and removed before the multiplication operation, and the abnormal value detection is based on statistical principles, for example, points deviating from the sequence mean by more than three times the standard deviation are considered abnormal. In some embodiments, the weighting operation can consider the correlation in the time dimension, that is, the weights of adjacent time points will affect each other, by introducing a time decay factor to make the weight of recent data slightly higher than that of distant data, so as to reflect the time variation trend of system performance.

[0068] The wavelet transform denoising process is applied to the weighted efficiency value sequence. In specific implementations, the wavelet transform denoising process first selects a suitable wavelet basis function, the selection of the wavelet basis function taking into account the waveform characteristics and noise characteristics of the weighted efficiency value sequence, the number of wavelet transform decomposition layers being adaptively determined according to the sequence length and the sampling frequency, and generally decomposed into four to eight layers to balance the denoising effect and the computational complexity. The wavelet transform decomposes the weighted efficiency value sequence into approximation coefficients and detail coefficients, the approximation coefficients representing the low-frequency trend components of the sequence, and the detail coefficients representing the high-frequency noise components. The threshold processing module performs soft thresholding or hard thresholding on the detail coefficients, sets the coefficients less than the threshold to zero, retains the coefficients greater than the threshold, and automatically calculates the threshold size according to the noise level estimate. The wavelet reconstruction process uses the processed coefficients to reconstruct the signal, obtaining the denoised weighted efficiency value sequence. The wavelet transform denoising process can effectively separate the real changes and random noise in the signal, and is particularly suitable for the processing of non-stationary time series. In some embodiments, the parameters of the wavelet transform denoising can be optimized by analyzing the signal-to-noise ratio of the sequence, and the signal-to-noise ratio is estimated based on the statistical characteristics of the sequence to ensure that the denoising process neither excessively smooths the real changes nor retains too much noise.

[0069] The Kalman filter is used for smoothing estimation. In specific implementations, the Kalman filter is modeled as a linear dynamic system, with the state variable being the real performance evaluation value and the observation variable being the weighted efficiency value after wavelet denoising. The state equation of the Kalman filter describes the variation of the performance evaluation value over time, and a first-order or second-order autoregressive model is usually used. The observation equation establishes the relationship between the state variable and the observation variable, assuming a linear relationship plus observation noise. The process noise covariance is adaptively adjusted according to the fluctuation amplitude of the operating parameters, and the fluctuation amplitude is obtained by calculating the variance or mean absolute deviation of the operating parameter sequence. When the operating parameters fluctuate greatly, the process noise covariance is increased to represent the increase in system uncertainty, and vice versa. The prediction step of the Kalman filter estimates the state value at the current time based on the state equation, and the update step corrects the predicted value using the observation value to produce the optimal smoothing estimate. The output of the Kalman filter is the denoised and smoothed performance evaluation value sequence, which retains the main trend characteristics of the original data while eliminating random fluctuations. Optionally, the Kalman filter can use an extended version to handle nonlinear relationships, or use an unscented Kalman filter to improve the estimation accuracy in nonlinear situations. The final output performance evaluation value sequence is stored in the performance knowledge base for use by the subsequent time series analysis module, and each typical operating condition category has a corresponding performance evaluation value sequence, which is updated regularly to reflect the long-term changes in device performance.

[0070] Referring to Figure 4The graph is a scatter plot with the horizontal axis representing the fan speed and the vertical axis representing the valve opening. The color mapping represents the heat recovery efficiency, with a gradient from deep purple to bright yellow visually presenting the efficiency gradient. This graph aims to visually analyze the correlation between heat recovery efficiency and key operating parameters. From the graph, one can observe the distribution characteristics of heat recovery efficiency under different fan speed and valve opening combinations. This correlation analysis is a key visualization support for extracting historical heat recovery efficiency data sequences and operating parameter sequences, calculating performance evaluation value sequences through statistical analysis, providing an intuitive working condition feature reference for subsequent weighted analysis based on operating parameter deviation and denoising smoothing processing, helping to identify the efficiency-parameter optimal interval under typical working condition categories, and then optimizing the operation control strategy of waste heat recovery equipment, and improving the accuracy and stability of heat recovery performance.

[0071] In a specific implementation, the performance evaluation value sequence is a smoothed numerical sequence obtained from typical working condition category analysis, with each number representing the heat recovery performance evaluation result at a specific time point. The sequence is arranged at fixed time intervals, usually one minute or five minutes, depending on the data acquisition frequency. The time series analysis module runs as an independent software component, receiving the performance evaluation value sequence from the upstream processing unit. The sequence data is transmitted through memory sharing or message queue to ensure data integrity and real-time performance. The difference operation uses a first-order backward difference formula. The calculation process traverses each data point in the performance evaluation value sequence, starting from the second point. The value of the current point is subtracted from the value of the previous point, and the difference is divided by the time interval between the two points to obtain the instantaneous change rate at that time point. The change rate calculation considers the possible slight fluctuations in the time interval and uses the exact time stamp difference for calculation to avoid errors introduced by uneven sampling. The calculated change rate values form a new change rate sequence. The change rate sequence has the same number of time points as the original performance evaluation value sequence minus one. Positive values in the change rate sequence indicate performance improvement, negative values indicate performance decline, and zero values indicate stable performance. The time series analysis module has a data verification mechanism inside to check the validity of the input sequence and reject sequences containing null or abnormal values to ensure the reliability of the analysis basis.

[0072] The rate of change sequence is filtered by moving average to eliminate noise. In specific implementation, the moving average filter uses a sliding window algorithm, and the window size is set according to the sampling frequency of the sequence and the expected smoothing degree. The common window size is five to eleven points. The moving average filter calculates the arithmetic mean of all rate of change values in the window as the filtered value of the center point of the window. The window slides with the calculation point and processes all data points in the rate of change sequence one by one. The filtering process uses special processing for data points at both ends of the sequence. For points at the beginning and end of the sequence that are less than half the window size, the average value is calculated by reducing the window or mirroring the extension to ensure that the length of the filtered sequence remains unchanged. The moving average filter effectively suppresses high-frequency random fluctuations in the rate of change sequence, which may be caused by measurement noise or short-term working condition disturbances, while preserving the main trend characteristics of the sequence. The filtered rate of change sequence is stored in a temporary buffer for subsequent trend segment extraction. The window size of the moving average filter can be adjusted adaptively. The adjustment is based on the noise level estimate of the sequence, which is obtained by calculating the local variance of the sequence. When a higher noise level is detected, the window size is automatically increased to improve the smoothing effect. It can be understood that the moving average filter is a linear filter, which may introduce a certain phase delay, but for offline or quasi-real-time analysis applications, this delay is within an acceptable range.

[0073] The rising segments and falling segments are extracted from the rate of change sequence. In specific implementation, the rising segment is defined as an interval where the rate of change at consecutive time points is all positive, and the falling segment is defined as an interval where the rate of change at consecutive time points is all negative. The segment extraction algorithm traverses the filtered rate of change sequence, identifies the turning points where the sign of the rate of change changes, and the continuous same sign interval between the turning points is a candidate segment. Each candidate segment needs to meet the minimum length requirement, which is usually set to three consecutive points to avoid misjudging isolated fluctuations as trend segments. The segment extraction process records the start index, end index, number of data points contained, and statistical characteristics of the rate of change within each segment. The extracted rising segments and falling segments are stored in different sets, and each segment object contains complete metadata information for subsequent index calculation. The segment extraction algorithm has special processing for boundary conditions. When the sequence starts or ends in an upward or downward state, the same sign interval starting from the sequence endpoint is also considered as a valid segment. It can be understood that segment extraction is the basis of trend analysis, and accurate identification of rising segments and falling segments is crucial for understanding the dynamics of efficiency changes.

[0074] The average rate of change of the ascending segments is calculated as the heat recovery trend indicator. In specific implementations, the average rate of change is calculated for each individual ascending segment, and the arithmetic mean of the rate of change values of all points in the segment is taken to obtain a scalar value representing the overall ascending intensity of the segment. If there are multiple ascending segments, the overall average of the average rates of change of all ascending segments is calculated as the final heat recovery trend indicator. The heat recovery trend indicator is a dimensionless value, and the value size reflects the rate of improvement of heat recovery performance. A larger positive value indicates a rapid improvement, and a smaller positive value indicates a slow improvement. The duration of the segment is considered in the calculation of the heat recovery trend indicator, and segments with longer duration have higher weights in the overall average to reflect the importance of long-term trends. The heat recovery trend indicator is output together with its statistical confidence interval, which is calculated based on the variance of the rate of change in the segment and represents the reliability of the indicator value. In some embodiments, the heat recovery trend indicator can be further decomposed into components at different time scales, such as a short-term fluctuation component and a long-term trend component, to provide more detailed trend information through multi-resolution analysis.

[0075] The average rate of change of the descending segments is calculated as the heat energy loss trend indicator. In specific implementations, the average rate of change of the descending segments is calculated similarly to the ascending segments, and the arithmetic mean of the rate of change values in each descending segment is taken, with the negative sign preserved to distinguish the descending direction. If there are multiple descending segments, the overall average of the average rates of change of all descending segments is calculated as the final heat energy loss trend indicator. The heat energy loss trend indicator is usually negative, and the absolute value size reflects the rate of deterioration of heat energy loss. A larger absolute value indicates a faster loss, and a smaller absolute value indicates a slower loss. The duration of the segment and the change amplitude are also considered in the heat energy loss trend indicator, which comprehensively reflects the severity of performance deterioration. The heat recovery trend indicator and the heat energy loss trend indicator together constitute the state description of the system performance, providing quantitative basis for control parameter generation. It can be understood that in some operating states, only ascending segments or only descending segments may be identified, in which case the missing indicator is replaced by a zero value or a historical average value to ensure that the system always has complete trend information available.

[0076] The control parameters are generated according to the calibrated heat recovery rate and the calibrated heat loss rate. In specific implementations, the calibrated heat recovery rate and the calibrated heat loss rate are numerical inputs received from the upstream correction module, representing the trend indicators after the working condition difference calibration. The control parameter generation module has a pre-set mapping relationship rule inside, which converts continuous rate values into discrete or continuous control set values. The calibrated heat recovery rate is mapped to the fan speed set value, and the mapping process uses a lookup table method or a function calculation method. The lookup table method is based on a pre-set mapping table for lookup, which contains the corresponding relationship between the rate interval and the corresponding speed set value. The function calculation method uses linear or non-linear functions to directly calculate, such as a proportional relationship or a segmented function with a dead zone. The calibrated heat loss rate is mapped to the valve opening set value, and the mapping relationship is usually an inverse relationship, i.e. the higher the loss rate, the smaller the opening set value, to suppress heat loss. The generated control parameters include specific fan speed set values and valve opening set values, with engineering units and effective range limits to ensure safe operation within the device range. The control parameter generation module checks the reasonableness of the set values to avoid producing instructions with sharp changes or exceeding the device capacity, and adds a gradual transition logic if necessary. It can be understood that control parameter generation is a key link between analysis results and execution actions, and its accuracy and reliability directly affect the control performance of the waste heat recovery system.

[0077] The control signal containing the fan speed set value and the valve opening set value is generated, and in specific implementations, the control signal is packaged in the format of a standard industrial communication protocol, such as a PROFIBUS DP message or a Modbus TCP data frame. The control signal contains a target device address field, a set value data field, a command type field and a timestamp field to ensure that the actuator can accurately identify and execute. The fan speed set value is represented in integer form, usually with revolutions per minute as the unit, and the numerical range corresponds to the acceptable input range of the frequency converter. The valve opening set value is represented in percentage form, with zero corresponding to the fully closed position and one hundred corresponding to the fully open position. The control signal is sent to the actuator of the waste heat recovery device through the industrial network, and the sending process has a retransmission mechanism and a response timeout detection to ensure that the instruction is reliably delivered. After receiving the control signal, the actuator parses the content and drives the motor or hydraulic device to adjust the fan speed and valve opening, realizing real-time adjustment of the operating state. The control signal generation period is synchronized with the system sampling period, usually once every one to several seconds, and the period is appropriately lengthened when the working condition is stable to reduce the frequency of device action. Optionally, the control signal can contain a feedforward compensation component, which is calculated based on the working condition prediction model to adjust the set value in advance to cope with expected disturbances, improving the response speed and stability of the control system. Referring to Table 1, a typical mapping relationship between the calibrated heat recovery rate and the fan speed set value is shown.

[0078] Table 1: Mapping relationship between calibrated heat recovery rate and fan speed set value

[0079] Calibrated heat recovery rate interval (unit: efficiency unit / min) Fan speed set value (unit: revolution per minute) Control action type (-∞,-0.5) 800 Emergency speed reduction [-0.5,0) 1000 Conservative operation [0,0.3) 1200 Maintenance operation [0.3,0.6) 1400 Aggressive operation [0.6,+∞) 1600 Maximum capacity operation

[0080] It can be understood that the specific numerical interval and set value of the mapping relationship table need to be adjusted on site according to the specific equipment model and process requirements, and optimized and adjusted according to the actual effect during the system operation. After the control signal is sent to the actuator of the waste heat recovery equipment, the actuator adjusts the fan speed and valve opening according to the set value, so as to change the working state of the heat exchanger and realize the accurate control of the flue gas waste heat recovery process.

[0081] Referring to Figure 5 The figure is a line chart, the horizontal axis is the running time, and the vertical axis is the trend index, which quantifies the change rate of heat recovery efficiency and heat energy loss. The purple line represents the heat recovery trend index, and the orange line represents the heat energy loss trend index. The figure is the core output of the time series analysis module. Through the first-order backward difference operation on the performance evaluation value sequence and the moving average filtering, the heat recovery trend index and the heat energy loss trend index are obtained. As can be seen from the figure, the heat recovery trend index shows a multi-period fluctuation characteristic, and the peak value interval reflects the stage of rapid improvement of heat recovery efficiency; the heat energy loss trend index is mostly in the negative value interval, and the absolute value change reflects the dynamic fluctuation of the loss rate. These indexes are the key basis for subsequent dynamic correction of heat recovery / loss rate and generation of control parameters. By real-time capturing the change law of the trend, the system can accurately adjust the running state of the waste heat recovery equipment, strengthen the recovery strategy when the heat recovery potential is large, and timely suppress the loss when the loss intensifies, so as to finally realize the dynamic optimization and performance maximization of the waste heat recovery process.

[0082] In a specific implementation, the current working condition feature set is a vector containing 5 parameter values, for example, the data collected in a specific running instance is temperature 280 degrees Celsius, pressure 1200 Pascal, flow rate 3500 cubic meters per hour, power output percentage 85%, and ambient humidity percentage 60%. The reference feature of the typical working condition category comes from the category center vector obtained from historical data clustering analysis, for example, the reference feature of category A can be temperature 275 degrees Celsius, pressure 1180 Pascal, flow rate 3480 cubic meters per hour, power output percentage 82%, and ambient humidity percentage 58%. The similarity calculation process needs to map the current working condition feature set vector and the reference feature vector of each category to the Riemannian manifold space model, which is a mathematical space that can describe the intrinsic geometric structure of the feature vector. In a specific implementation, the construction of the Riemannian manifold space model is based on the covariance matrix properties of the feature vector, and each feature vector is regarded as a point on the manifold, and the distance metric on the manifold uses geodesic distance instead of Euclidean distance, which can better reflect the nonlinear relationship between features. When calculating the geodesic distance between the current working condition feature point and the reference feature point of each category, it is necessary to solve the shortest path length connecting the two points on the manifold, which involves complex differential geometry calculations.

[0083] The Riemannian manifold space model of the feature vector is constructed, and in a specific implementation, the Riemannian manifold space model regards the set of feature vectors as a differential manifold, and each point on the manifold corresponds to a feature vector. The geometric structure of the manifold is determined by the statistical relationship between the feature parameters. For a feature vector containing 5 parameters, the Riemannian manifold space model can be constructed as a 5-dimensional manifold, and the manifold is equipped with a Riemannian metric tensor that defines the inner product operation rule of the tangent space at any point on the manifold. The specific form of the Riemannian metric tensor can be determined by the covariance matrix of the feature vector, which reflects the variation relationship and scale difference between different feature parameters. In actual calculations, the current working condition feature set vector and the reference feature vector of each category need to be represented as point coordinates on the manifold through a mapping relationship, and this mapping process needs to keep the geometric properties of the feature vector unchanged.

[0084] The geodesic distance between the current operating condition feature point and the reference feature point of each category is calculated. In specific implementations, the calculation of the geodesic distance requires solving the shortest path connecting two points on the Riemannian manifold, which is called the geodesic. Taking the current operating condition feature point and a reference feature point of a certain category as an example, assuming that the current operating condition feature point corresponds to point P on the manifold, and the reference feature point corresponds to point Q, the geodesic connecting point P and point Q needs to be found and its length needs to be calculated. The solution of the geodesic usually needs to be realized through exponential mapping and logarithmic mapping. The logarithmic mapping maps point Q into the tangent space at point P to obtain a tangent vector, and the norm of this tangent vector is the geodesic distance. The calculation process involves numerical calculation of Riemannian exponential mapping and Riemannian logarithmic mapping, and iterative solution of the geodesic equation is required. The calculation result of the geodesic distance is a scalar value, which represents the real distance between the current operating condition feature point and the reference feature point of the category on the manifold. The smaller the distance value, the higher the similarity. Taking the category corresponding to the minimum distance as the reference category means that the most similar historical operating mode to the current operating condition is selected from all typical operating condition categories.

[0085] A tangent space is established at the current operating condition feature point. In specific implementations, the tangent space is a linear approximation space of the Riemannian manifold at a certain point, which is a Euclidean space. Establishing the tangent space requires determining the tangent plane at the current operating condition feature point, which is spanned by the tangent vectors at the point. The tangent vector represents the velocity vector of the curve passing through the point on the manifold at the point. The establishment process of the tangent space includes selecting the origin of the tangent space (i.e. the current operating condition feature point itself) and determining the basis vectors of the tangent space, which can obtain a set of standard orthogonal bases through the Schmidt orthogonalization process. The dimension of the tangent space is the same as that of the manifold. For a 5-dimensional manifold, the tangent space is also a 5-dimensional Euclidean space. The reference category reference feature point is projected into the tangent space, which is realized through Riemannian logarithmic mapping. The Riemannian logarithmic mapping maps a point on the manifold into a vector in the tangent space, which is called the projection vector. The direction of the projection vector represents the change direction of the reference category reference feature point relative to the current operating condition feature point, and the length of the projection vector represents the difference between the two. It can be understood that the establishment of the tangent space provides a convenient working space for subsequent vector decomposition and correction calculation, and converts complex manifold operations into relatively simple linear algebra operations.

[0086] The projection vector is decomposed into a radial component and a tangential component. In specific implementations, the projection vector is a 5-dimensional vector in the tangent space, and the decomposition operation requires a decomposition reference direction to be determined. The radial component represents the projection of the projection vector along a specific direction, which is usually chosen as the direction from the current operating condition feature point to a reference point on the manifold, or the direction of variation of a dominant parameter in the feature vector. The tangential component is the remaining part of the projection vector after subtracting the radial component, representing the projection in the direction perpendicular to the radial direction. The magnitude of the radial component reflects the difference in feature intensity between the current operating condition and the reference category, such as the difference in parameter values such as temperature, pressure, etc. The direction of the tangential component reflects the difference in operating mode between the current operating condition and the reference category, such as the change in the ratio or coupling relationship between different parameters. The decomposition calculation is realized through the vector projection formula, which first calculates the scalar projection of the projection vector on the unit vector of the radial direction, then multiplies it by the unit vector of the radial direction to obtain the radial component, and finally subtracts the radial component from the original projection vector to obtain the tangential component. In some embodiments, the direction of the radial component can be determined by the principal component analysis method, and the direction with the largest feature variation is chosen as the radial direction.

[0087] The intensity correction factor is calculated according to the magnitude of the radial component. In specific implementations, the intensity correction factor is a scalar coefficient used to adjust the amplitude of the trend indicator. The calculation of the intensity correction factor is based on the norm of the radial component, i.e. the length of the radial vector. The larger the length of the radial vector, the greater the difference in feature intensity between the current operating condition and the reference category. The intensity correction factor can be designed as a function of the norm of the radial component, such as a linear function, an exponential function, or a piecewise function, and the specific form needs to be determined according to the actual application scenario. The value range of the intensity correction factor is usually set between 0 and 2, with a value of 1 indicating no correction, a value less than 1 indicating a need to weaken the trend indicator, and a value greater than 1 indicating a need to enhance the trend indicator. When calculating the intensity correction factor, the direction of the radial component also needs to be considered. If the radial component points in the direction of feature enhancement, the intensity correction factor may be greater than 1, and vice versa. It can be understood that the introduction of the intensity correction factor enables the trend indicator to adapt to the difference in variation amplitude under different intensity conditions, improving the adaptability of control. Optionally, the calculation of the intensity correction factor can introduce smoothing processing to avoid drastic changes in the correction factor due to small fluctuations in the norm of the radial component.

[0088] The mode correction matrix is calculated based on the direction of the tangential component, which reflects the difference in operating mode between the current operating condition and the reference category. The mode correction matrix can be determined by the relationship between the tangential component and other reference directions, for example, calculating the cosine value of the angle between the tangential component and a set of basis vectors as the matrix elements. The design of the mode correction matrix needs to ensure its invertibility in order to subsequently perform inverse transformation on the heat loss trend index. The element values of the mode correction matrix reflect the adjustment weight of the influence of different characteristic parameters on the trend index. The diagonal elements represent the direct influence adjustment of the respective parameters, and the non-diagonal elements represent the adjustment of the cross-influence between parameters. The calculation of the mode correction matrix may involve matrix operations such as eigenvalue decomposition or singular value decomposition to extract the main direction features of the tangential component. Optionally, the mode correction matrix can be designed as a diagonally dominant matrix to ensure the stability and reasonableness of the transformation.

[0089] The heat recovery trend index is multiplied by the intensity correction factor and then linearly transformed by the mode correction matrix. In specific implementation, the heat recovery trend index is a scalar value representing the change trend of heat recovery efficiency. First, the heat recovery trend index is multiplied by the intensity correction factor to obtain an intermediate value after intensity correction. Then, this intermediate value is regarded as a one-dimensional vector and multiplied by the mode correction matrix. Since the heat recovery trend index is a scalar, the mode correction matrix needs to be adjusted accordingly to the form of a one-dimensional row vector or column vector. The result of linear transformation is a new scalar value, called calibrated heat recovery rate, which takes into account the differences in intensity and mode between the current operating condition and the reference category, and is more suitable for the current actual operating conditions. The physical meaning of the calibrated heat recovery rate is the same as that of the original heat recovery trend index, but the numerical value has been adjusted accordingly. It can be understood that this calibration process enables the system to automatically adjust the control strategy according to different operating condition characteristics, improving the adaptability and efficiency of the waste heat recovery system.

[0090] The heat loss trend index is multiplied by the intensity correction factor, and then linearly transformed by the inverse of the pattern correction matrix. In a specific implementation, the heat loss trend index is also a scalar value, representing the change trend of the heat loss. Similar to the heat recovery trend index, the heat loss trend index is first multiplied by the intensity correction factor to obtain an intermediate value after intensity correction. Then the intermediate value is linearly transformed by the inverse of the pattern correction matrix, which ensures the reversibility of the transformation, and makes the calibration direction of the heat loss trend index opposite to that of the heat recovery trend index. The result of the linear transformation is referred to as the calibrated heat loss rate, which reflects the adjusted heat loss change rate according to the current operating characteristics. The calibrated heat recovery rate and the calibrated heat loss rate are used as the final calibration output to generate the control parameters of the waste heat recovery device. Optionally, the calibration calculation process can introduce boundary limits to ensure that the calibrated rate values are within a reasonable physical range, avoiding the generation of unrealistic control instructions.

[0091] While embodiments of the present application have been shown and described, it is to be understood that the embodiments described are merely divergences of the principles and spirit of the present application, and that many modifications, substitutions, replacements and variations can be made by those of ordinary skill in the art without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for recovering waste heat from the flue gas of a setting machine, characterized in that, The method comprises: The temperature, pressure and flow rate of the flue gas are continuously acquired by a sensor array installed on the flue gas duct of the setting machine, while the real-time power output of the setting machine and the ambient humidity data are recorded to form a current working condition feature set; Using the current working condition feature set, search for historical operation records with similar features in the historical database, and perform pattern recognition processing on these historical operation records to divide into multiple typical working condition categories; for each typical working condition category, extract the corresponding historical heat recovery efficiency data sequence and operation parameter sequence, and calculate the performance evaluation value sequence under each category by statistical analysis; The performance evaluation value sequence is input into a time series analysis module to output heat recovery trend indicators and heat energy loss trend indicators; similarity calculation is performed between the current working condition feature set and the reference features of each typical working condition category, the category with the highest similarity is selected as the reference category, and the heat recovery trend indicators and heat energy loss trend indicators are dynamically corrected according to the difference between the current working condition feature set and the reference category reference features to obtain the calibrated heat recovery rate and the calibrated heat energy loss rate; According to the calibrated heat recovery rate and the calibrated heat energy loss rate, control parameters are generated to adjust the operating state of the waste heat recovery equipment.

2. The method for recovering waste heat from the flue gas of a setting machine according to claim 1, characterized in that, The temperature, pressure and flow rate of the flue gas are continuously acquired by a sensor array installed on the flue gas duct of the setting machine, while the real-time power output of the setting machine and the ambient humidity data are recorded to form a current working condition feature set, which specifically comprises: The instantaneous temperature value at the flue gas inlet is measured using a temperature sensor, the real-time pressure value in the flue gas duct is measured using a pressure sensor, the flue gas volume flow value is monitored using a flow meter, the power output percentage is read from the setting machine controller, and the humidity percentage value is obtained from the environment sensor; the temperature value, pressure value, flow value, power output percentage and humidity percentage value are aligned by timestamp, combined into a multi-dimensional vector, and constitute the current working condition feature set.

3. The method for recovering waste heat from the flue gas of a setting machine according to claim 1, characterized in that, Using the current working condition feature set, search for historical operation records with similar features in the historical database, and perform pattern recognition processing on these historical operation records to divide into multiple typical working condition categories, which specifically comprises: Calculate the Euclidean distance between the current working condition feature set and the feature vector of each historical operation record in the historical database, and select the historical operation records with a distance less than a threshold value as an initial similar set; all historical operation records in the initial similar set are grouped using a clustering algorithm, and the clustering is based on the combined features of temperature readings, pressure readings and power output values to form multiple typical working condition categories.

4. The method for recovering waste heat from the flue gas of a setting machine according to claim 1, characterized in that, For each typical working condition category, extract the corresponding historical heat recovery efficiency data sequence and operation parameter sequence, and calculate the performance evaluation value sequence under each category by statistical analysis, which specifically comprises: Retrieving historical heat recovery efficiency values from member records of each typical working condition category and arranging them in chronological order to form a historical heat recovery efficiency data sequence; meanwhile, extracting operating parameter values at corresponding time points, including fan speed and valve opening, to form an operating parameter sequence; performing weighted average calculation on the historical heat recovery efficiency data sequence and the operating parameter sequence, with weights determined based on deviations of operating parameters from standard values, and outputting a performance evaluation value sequence.

5. The method for recovering waste heat from the flue gas of a setting machine according to claim 4, characterized in that, The weighted average calculation on the historical heat recovery efficiency data sequence and the operating parameter sequence, with weights determined based on deviations of operating parameters from standard values, outputs a performance evaluation value sequence, specifically including: Based on deviations of the operating parameter sequence from preset standard value ranges, determining dynamic weight coefficients corresponding to each operating parameter through an adaptive weight calculation mechanism; Multiplying each efficiency value in the historical heat recovery efficiency data sequence by a dynamic weight coefficient at the corresponding time point to obtain a weighted efficiency value sequence; Performing denoising processing on the weighted efficiency value sequence to eliminate interference components, and then performing smoothing estimation to optimize data stability, finally outputting a performance evaluation value sequence.

6. The method for recovering waste heat from the flue gas of a setting machine according to claim 1, characterized in that, The time series analysis module is configured to perform: Performing difference operation on the performance evaluation value sequence to calculate the change rate at adjacent time points, obtaining a change rate sequence; extracting the rising section and the falling section from the change rate sequence, the average change rate of the rising section as a heat recovery trend indicator, and the average change rate of the falling section as a heat energy loss trend indicator; Wherein the difference operation adopts a first-order backward difference formula, and the change rate sequence is filtered by moving average to eliminate noise.

7. The method for recovering waste heat from the flue gas of a setting machine according to claim 1, characterized in that, The difference between the current working condition feature set and the reference category benchmark feature is dynamically corrected to the heat recovery trend indicator and the heat energy loss trend indicator to obtain a calibrated heat recovery rate and a calibrated heat energy loss rate, specifically including: Constructing a geometric space model of feature vectors, mapping the current working condition feature set and the benchmark features of each category to points on a manifold; Calculating the spatial distance between the current working condition feature point and the benchmark feature points of each category, and selecting the category with the smallest distance as the reference category; Establishing a linear approximation space corresponding to the current working condition feature point, projecting the reference category benchmark feature point to the linear approximation space to obtain a projection vector; Performing component decomposition on the projection vector to obtain an intensity component representing the intensity difference and a mode component representing the operation mode difference; Determining an intensity correction factor based on the intensity component and a mode correction parameter based on the mode component; Combining the intensity correction factor and the mode correction parameter to correct the heat recovery trend indicator to obtain the calibrated heat recovery rate; Combining the reverse adaptation parameters corresponding to the intensity correction factor and the mode correction parameter to correct the heat energy loss trend indicator to obtain the calibrated heat energy loss rate.

8. The method for recovering waste heat from the flue gas of a setting machine according to claim 1, characterized in that, Generating control parameters according to the calibrated heat recovery rate and the calibrated heat energy loss rate to adjust the operating state of the waste heat recovery equipment, specifically including: mapping the calibrated heat recovery rate to a fan speed set value and the calibrated heat energy loss rate to a valve opening set value; generating a control signal containing the fan speed set value and the valve opening set value and sending it to the actuator of the waste heat recovery equipment.

9. The method for recovering waste heat from the flue gas of a setting machine according to claim 3, characterized in that, All historical run records in the initial similarity set are grouped using a clustering algorithm, the clustering being based on a combination of temperature readings, pressure readings, and power output values, wherein the clustering algorithm uses a mean shift method to automatically determine the number of classes.

10. A setting machine flue gas waste heat recovery system based on a memory, a processor and a computer program stored in the memory and running on the processor, characterized in that, The processor, when executing the computer program, implements the steps of the method for recovering waste heat from the flue gas of a stenter according to any one of claims 1 to 9.

Images