Method and apparatus for plasma treatment of contaminated soil based on active perturbation response

Abstract

The application discloses a method and device for plasma treatment of contaminated soil based on active disturbance response. The method first establishes a baseline discharge state for the current batch, and when an abnormal discharge characteristic is detected, at least two of the following are performed: air gap micro-disturbance, airflow micro-disturbance, mechanical micro-disturbance, low-energy probe discharge, and stop and recovery diagnosis. The voltage, current, and charge waveforms before and after each active diagnostic action are collected, and the corresponding waveform tensor, differential response characteristic, and physical response index are constructed. The above data is input into a pre-trained multi-disturbance branch hierarchical classification model, which outputs the first and second categories of impurities, risk level, and classification confidence. Based on the identification results, a treatment strategy is generated and executed, and active diagnosis is performed again after treatment to verify the treatment effect. The application can distinguish different impurity disturbance sources with similar natural discharge waveforms, improving the stability and safety of plasma treatment of complex heterogeneous contaminated soil.

Description

Technical Field

[0001] This application belongs to the field of environmental pollution control, and in particular relates to a method and apparatus for treating contaminated soil using plasma based on active disturbance response. Background Technology

[0002] Dielectric barrier discharge (DBD) plasma can generate high-energy electrons, ozone, hydroxyl radicals, oxygen atoms and other active species under normal or near-normal pressure conditions. It has been used for the degradation of organic pollutants, waste gas treatment, water treatment and remediation of organically contaminated soil.

[0003] However, in actual processes of treating contaminated soil or oil-based drill cuttings, the objects being treated are usually not homogeneous media, but rather contain a large number of heterogeneous components. For example, petroleum hydrocarbon-contaminated soil may contain leaves, roots, wood chips, plastic sheets, pebbles, glass, sand, metal fragments, salt aggregates, and undispersed soil clods; oil-based drill cuttings may also contain coarse rock cuttings, agglomerated mud lumps, salt-containing conductive points, and other foreign impurities. These impurities can cause problems such as local electric field distortion in the discharge gap, changes in equivalent capacitance, premature or delayed local breakdown, abnormal number and amplitude of micro-discharge pulses, insufficient coverage of the discharge area, local hot spots, arc risk, and uneven treatment.

[0004] DBD plasma reactors (such as CN120115513A) typically include a high-voltage electrode, a grounding electrode, a dielectric barrier layer, a sample-bearing area, a gas inlet, and a tail gas outlet, used for the discharge treatment of contaminated soil or oil-based drill cuttings. Existing DBD treatment systems generally focus primarily on macroscopic indicators such as discharge power, equivalent capacitance, temperature, and degradation rate, lacking the ability to identify the types of internal impurity disturbances within the treated material online. Even if operators can observe waveform anomalies on an oscilloscope, they often can only rely on experience to reduce voltage, shut down the system, or stir the material, unable to take targeted measures based on different impurity types. Existing oscilloscope waveform analysis methods are typically used to determine changes in discharge power, breakdown state, discharge mode, or equivalent capacitance, primarily analyzing the electrical state of the discharge system itself. For heterogeneous solid materials such as contaminated soil or oil-based drill cuttings, even abnormalities in voltage, current, or Lissajous figures can be caused by a variety of different reasons. For example, conductive particles, carbonized leaf spots, and localized wet clumps may all exhibit decreased discharge voltage and increased peak current; insulating hard impurities and large soil clods may exhibit delayed discharge, reduced pulse number, and increased discharge energy fluctuations; lightweight paper, plastic film, and leaf fragments may exhibit intermittent discharge anomalies. Relying solely on the passive waveform characteristics under natural discharge conditions is insufficient to reliably distinguish the sources of these material disturbances.

[0005] Therefore, existing technologies lack an active diagnostic method for contaminated soil or oil-based drill cuttings disturbances. Summary of the Invention

[0006] Therefore, the purpose of this application is to provide a plasma treatment method and apparatus for contaminated soil based on active perturbation response, which aims to identify different impurities in contaminated soil and improve the treatment effect of contaminated soil.

[0007] In a first aspect, this application provides a plasma treatment method for contaminated soil based on active perturbation response, comprising: A low-energy detection discharge was performed on the current batch of contaminated soil to be treated, and reference voltage waveform, reference current waveform and reference charge waveform were collected to establish the reference discharge characteristics of the current batch. During the plasma treatment of contaminated soil, the discharge state is monitored. If the discharge state has an abnormal discharge response relative to the current batch reference discharge characteristics, at least two different active diagnostic actions are executed according to a preset active diagnostic sequence. The active diagnostic actions include air gap perturbation action, airflow perturbation action, mechanical perturbation action, low-energy detection discharge action, and parking recovery action. Data acquisition steps: Collect voltage waveforms, current waveforms, and charge waveforms before and after each active diagnostic action, and then construct waveform tensors, differential response characteristics, and physical response exponents; The waveform tensor, differential response features, physical response index, and operating parameters corresponding to each active diagnostic action are input into a pre-trained multi-perturbation branch hierarchical classification model to obtain the first-level category, second-level category, risk level, and classification confidence of the impurities. If the classification confidence level does not meet the preset conditions, a new active diagnostic action is selected based on the probability distribution of the current categories, and the data acquisition step is returned. If the classification confidence level meets the preset conditions, a treatment strategy is generated based on the primary category, secondary category, and risk level of the impurities. After the treatment strategy is executed, when plasma treatment is performed on the contaminated soil, at least one active diagnostic action is executed again, and voltage waveforms, current waveforms, and charge waveforms before and after the execution of the treatment strategy are collected to determine the degree of abnormal response after the execution of the treatment strategy. If the degree of abnormal response is higher than the preset range, the treatment level is upgraded, and another treatment strategy is selected or a safety interlock is triggered.

[0008] Optionally, the air gap perturbation action includes briefly increasing or decreasing the discharge air gap by a preset distance based on the reference air gap; The airflow perturbation action includes changing the flow rate, flow direction, or pulsation mode of the processed gas; The mechanical perturbation actions include turning, vibrating, compacting, or partially breaking up the contaminated soil. The low-energy detection discharge action includes reducing the discharge power, peak voltage, duty cycle, or pulse duration to a preset ratio of the parameters corresponding to the normal processing stage, while maintaining the ability to generate repeatable micro-discharge pulses. The parking recovery action includes pausing the high-voltage output for a preset time and then restarting the discharge, and collecting the recovery response during the restarting process.

[0009] Optionally, the waveform tensor includes voltage, current, and charge waveforms before the active diagnostic action is executed, voltage, current, and charge waveforms after the action is executed, and differential signals between the waveforms before and after the action. The differential response characteristics include at least three of the following: discharge voltage difference, micro-discharge pulse number difference, pulse peak current difference, long-tail current difference, Lissajous figure area difference, equivalent capacitance difference, pulse phase difference, single-cycle energy difference, high-frequency energy ratio difference, and pulse cluster number difference. The physical response index includes at least two of the following: conductive channel retention index, shading migration index, compaction response index, organic self-evolution trajectory index, and drift correlation index. The drift correlation index includes airflow correlation index and tumbling correlation index. The airflow correlation index is determined based on the delay correlation between the abnormal waveform amplitude sequence and the gas flow rate change signal, airflow direction switching signal, or pulsed air intake signal; the tumbling correlation index is determined based on the delay correlation between the abnormal waveform amplitude sequence and the tumbling action, vibration action, or other mechanical disturbance signal. Operating parameters include at least three of the following: discharge power, frequency, duty cycle, air gap, gas flow rate, sample layer thickness, and sample water content.

[0010] Optionally, the multi-perturbation branch hierarchical classification model includes: The waveform deep feature extraction branch is used to extract waveform deep features based on the waveform tensor of the active diagnostic action; The physical feature branch is used to extract physical feature representations based on differential response characteristics, physical response index, and operating parameters. The attention fusion module is used to obtain attention intermediate features from the waveform deep features, differential response features, and operating condition features of each active diagnostic action; it receives the active diagnostic action mask and determines the fusion weight of each type of active diagnostic action based on the attention intermediate features and the active diagnostic action mask; then, it uses the fusion weight to perform weighted fusion of the attention intermediate features of several active diagnostic actions to obtain the active disturbance waveform features. The first-level classification head is used to determine the physical prior classification score based on the physical response index characteristics; determine the data-driven classification score based on the active perturbation waveform characteristics and physical feature representation; calculate the gating coefficient based on the waveform signal quality, the completeness of the active diagnostic action and the stability of the current batch baseline waveform, and fuse the physical prior classification score and the data-driven classification score based on the gating coefficient to determine the first-level category and first-level classification confidence of the impurity. The secondary classification head is used to determine the secondary category, confidence level, and risk level of impurities based on the waveform and physical characteristics of the active disturbance.

[0011] Optionally, the training process of the multi-perturbation branch hierarchical classification model includes: Perform at least two active diagnostic actions on contaminated soil samples containing known primary and secondary impurities, and acquire voltage, current, and charge waveforms before and after each active diagnostic action, and form waveform tensors, differential response characteristics, physical response exponents, and operating parameters as sample data; The sample data were divided into training, validation, and test sets according to the batches of contaminated soil. A network model was established, including a waveform deep feature extraction branch, a physical feature branch, a multi-perturbation attention fusion module, a first-level classification head, and a second-level classification head. The network model is trained using first-level classification loss, second-level classification loss, and physical prior consistency loss to obtain a multi-perturbation branch hierarchical classification model.

[0012] Optionally, the primary categories include conductive channel type impurities, insulating hard shield type impurities, large-size aggregate type impurities, and organic disturbance type impurities; When the primary category is conductive channel type impurities, the corresponding secondary categories include stable conductive particles, wet salt conductive clusters, and carbonized conductive points; When the primary category is insulating hard shielding impurities, the corresponding secondary categories include sheet-like hard insulators, block-like hard insulators, and dispersed particle enrichment areas; When the primary category is large-sized aggregated impurities, the corresponding secondary categories include ordinary soil lumps, bonded aggregates, and surface crusts. When the primary category is organic disturbance type impurities, the corresponding secondary categories include slow-time organic evolutions, airflow-related light drifts, and wood chips or root and stem organic matter; The model also outputs one of the following risk levels: general disturbance, moderate risk, and risk requiring interlocking measures.

[0013] Optionally, when the highest probability of the primary classification is lower than the first confidence threshold, or when the difference between the highest probability and the second highest probability of the primary classification is lower than the second confidence threshold, a mixed perturbation or uncertain category is output, and the next active diagnostic action is selected based on the difference between the candidate categories. When the classification probabilities of insulating hard-shielding impurities and large-size agglomerated impurities are similar, compaction or vibration diagnosis should be selected. When the classification probabilities of slow-time organic evolutions and airflow-related lightweight drifts are close, airflow perturbation or parking recovery diagnosis is selected.

[0014] Secondly, this application provides a plasma treatment device for contaminated soil based on active perturbation response, comprising: A dielectric barrier discharge reactor is used to hold contaminated soil to be treated. An adjustable high-voltage power supply is connected to the dielectric barrier discharge reactor to adjust the discharge power, frequency, duty cycle, or peak voltage. The waveform acquisition module is used to simultaneously acquire discharge voltage, current, and charge waveforms. The active diagnostic execution module includes at least two of the following: an air gap perturbation mechanism, an airflow control mechanism, a mechanical perturbation mechanism, and a parking recovery control unit; The disposal execution module includes at least one of the following: an air gap adjustment mechanism, an airflow adjustment mechanism, a turning and mixing mechanism, a compaction or vibration mechanism, a crushing or screening mechanism, a magnetic separation mechanism, and a safety interlock mechanism; The controller is connected to the adjustable high-voltage power supply, the waveform acquisition module, the active diagnostic execution module, and the treatment execution module, respectively. The controller is used to establish the baseline discharge characteristics of the current batch, control the active diagnosis execution module to perform at least two active diagnosis actions, construct differential response characteristics based on the waveforms before and after the active diagnosis actions, input the differential response characteristics into a pre-trained impurity identification model, and control the disposal execution module based on the impurity category, risk level and classification confidence level output by the model.

[0015] Thirdly, this application provides a contaminated soil treatment device, including the contaminated soil plasma treatment device based on active perturbation response as described above.

[0016] Fourthly, this application provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores at least one piece of program code, which is executed by a processor to implement the plasma treatment method for contaminated soil based on active perturbation response as described in any of the preceding claims.

[0017] The beneficial effects of the technical solution provided in this application include: This application provides a plasma treatment method for contaminated soil based on active perturbation response. By applying active diagnostic actions during plasma treatment and acquiring voltage, current, and charge waveforms before and after the perturbation, waveform tensors, differential response features, and physical response indices are constructed. Based on these waveform tensors, differential response features, and physical response indices, a pre-trained multi-perturbation branch hierarchical classification model is used to identify impurities, and different treatment strategies are formulated according to the impurity identification results. This method can effectively improve the treatment effect on contaminated soil. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A flowchart of a plasma treatment method for contaminated soil based on active perturbation response provided in an embodiment of this application; Figure 2 This is a schematic diagram illustrating the voltage, current, and charge waveform acquisition process during plasma processing, provided in an embodiment of this application. Figure 3 A schematic diagram of the waveforms of different types of impurities after a perturbation action is applied, provided in an embodiment of this application; Figure 4 A flowchart of a plasma treatment method for contaminated soil based on active perturbation response provided in another embodiment of this application; Figure 5 A schematic diagram illustrating different types of impurity adjustment strategies provided in an embodiment of this application; Figure 6 A structural block diagram of plasma treatment of contaminated soil based on active perturbation response provided in an embodiment of this application; Figure 7 This is a structural block diagram of a contaminated soil treatment device provided in one embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] Figure 1A method for identifying contaminated soil impurities during plasma processing, as provided in one embodiment of this application. See also... Figure 1 ,include: S101. Perform low-energy detection discharge on the current batch of contaminated soil to be treated, collect the reference voltage waveform, reference current waveform and reference charge waveform, and establish the reference discharge characteristics of the current batch.

[0022] S102. During the plasma treatment of contaminated soil, monitor the discharge state. If the discharge state has an abnormal discharge response relative to the current batch reference discharge characteristics, execute at least two different active diagnostic actions according to a preset active diagnostic sequence. The active diagnostic actions include air gap perturbation action, airflow perturbation action, mechanical perturbation action, low-energy detection discharge action, and parking recovery action.

[0023] S103. Acquisition Steps: Acquire voltage waveforms, current waveforms, and charge waveforms before and after each active diagnostic action, and then construct waveform tensors, differential response characteristics, and physical response exponents.

[0024] S104. Input the waveform tensor, differential response features, physical response index and operating parameters corresponding to each active diagnostic action into the pre-trained multi-perturbation branch hierarchical classification model to obtain the primary category, secondary category, risk level and classification confidence of the impurities.

[0025] S105. If the classification confidence level does not meet the preset conditions, select a new active diagnostic action based on the probability distribution of the current categories and return to the data acquisition step; if the classification confidence level meets the preset conditions, generate a treatment strategy based on the primary category, secondary category and risk level of the impurities; after executing the treatment strategy, when performing plasma treatment on the contaminated soil, execute at least one active diagnostic action again and collect the voltage waveform, current waveform and charge waveform before and after executing the treatment strategy to determine the degree of abnormal response after executing the treatment strategy; if the degree of abnormal response is higher than the preset range, upgrade the treatment level and reselect another treatment strategy or trigger a safety interlock.

[0026] In some embodiments, in step S101, before the formal treatment of the dielectric barrier discharge plasma, a low-energy detection discharge is performed on the current batch of contaminated soil to be treated, and a reference voltage waveform, a reference current waveform, and a reference charge waveform are collected, and a reference discharge characteristic of the current batch is established based on the reference waveform.

[0027] In some embodiments, voltage waveforms can be acquired using a high-voltage probe. Current waveforms can be acquired using a current probe. Charge waveform is collected by measuring capacitance. .

[0028] In some embodiments, a DBD plasma reactor is typically used to treat contaminated soil using plasma. A DBD plasma reactor generally includes a high-voltage electrode, a ground electrode, a dielectric barrier layer, a sample-bearing area, a gas inlet, and a tail gas outlet. In this application, the voltage waveform is obtained by acquiring the voltage waveform of the high-voltage electrode, the current waveform is obtained by acquiring the loop current between the high-voltage electrode and the ground electrode, and the charge waveform is obtained by acquiring the amount of charge accumulated in the loop between the high-voltage electrode and the ground electrode.

[0029] In some embodiments, voltage-current waveforms can also be acquired using an oscilloscope. And Lissajous figures .

[0030] In some embodiments, in step S102, the discharge state is monitored during normal plasma processing. When an abnormal discharge response relative to the current batch reference discharge characteristics is detected, at least two different active diagnostic actions are executed according to a preset active diagnostic sequence. The active diagnostic actions include air gap perturbation action, airflow perturbation action, mechanical perturbation action, low-energy detection discharge action, and parking recovery action.

[0031] In some embodiments, the process of each diagnostic action in the active diagnostic sequence in step S102 is as follows: The air gap perturbation action includes briefly increasing or decreasing the discharge air gap by a preset distance based on the reference air gap. The airflow perturbation action includes changing the flow rate, flow direction, or pulsation mode of the processed gas; The mechanical perturbation actions include turning, vibrating, compacting, or partially breaking up the contaminated soil. The low-energy detection discharge action includes reducing the discharge power, peak voltage, duty cycle, or pulse duration to a preset ratio of the parameters corresponding to the normal processing stage, while maintaining the ability to generate repeatable micro-discharge pulses. The parking recovery action includes pausing the high-voltage output for a preset time and then restarting the discharge, and collecting the recovery response during the restarting process.

[0032] See Figure 2 In some embodiments, the waveform acquisition process of this application may include the following stages: The first stage is the low-energy discharge stage.

[0033] The low-energy detection discharge stage, as referred to in this application, is a discharge state in which the discharge power, peak voltage, duty cycle, or duration of a single discharge is lower than that of the normal pollutant treatment stage, but repeatable and detectable micro-discharge pulses can still be formed in the discharge gap. The purpose of the low-energy detection discharge stage is to obtain the initial discharge response of the current batch of contaminated soil without significantly changing the current material morphology and impurity state, and to detect the risk of local conductive channels or quasi-arcs in advance.

[0034] In one embodiment, the low-energy detection discharge power is 10% to 50% of the normal processing power, or the low-energy detection discharge duty cycle is 10% to 50% of the normal processing duty cycle. The specific ratio is determined based on the reactor structure, air gap, and material state.

[0035] The low-energy detection discharge does not refer to the breakdown of the dielectric barrier layer; the dielectric barrier layer remains insulating, and the discharge occurs in the air gap region.

[0036] This application first collects voltage waveforms, current waveforms, charge waveforms, voltage-current waveforms, and Lissajous figures during the low-energy discharge phase to form a reference waveform.

[0037] After the low-energy discharge phase ends, the second phase, the high-energy discharge phase, begins. At this point, once the voltage exceeds the breakdown value, a large number of filamentary streamers appear instantaneously in the gap, and typical "spike-like" spikes appear on the current waveform, indicating a sudden increase in power.

[0038] In this application, the high-energy discharge stage is divided into several sub-stages based on the type of perturbation action applied during the high-energy discharge stage, namely: The first sub-stage is the air gap perturbation stage.

[0039] During the air gap perturbation stage, the air gap in the plasma processing (the air gap in a DBD plasma reactor generally refers to the region between dielectric barrier layers) is finely adjusted, such as by using a micro-push rod or lead screw mechanism to adjust the distance between the dielectric barrier layers, causing a short-term change in the air gap of 0.1 mm to 0.5 mm. Subsequently, the voltage, current, and charge waveforms after the air gap perturbation are acquired. Based on the voltage, current, and charge waveforms before and after the air gap perturbation, the differential response matrix before and after the air gap perturbation is obtained. .

[0040] Specifically, after obtaining the voltage waveform, current waveform, and charge waveform before and after the air gap perturbation, the waveform features in the voltage waveform, current waveform, and charge waveform before and after the air gap perturbation are extracted, and the extracted waveform features before and after the air gap perturbation are differentially divided to form a differential response matrix.

[0041] In some embodiments, the extracted waveform features include: Discharge voltage Micro-discharge pulse count Single-pulse peak current Duration of the long tail of the current Lissajous figure area Air gap equivalent capacitance Pulse phase center Periodic energy Total Harmonic Distortion High-frequency energy ratio The number of micro-discharge pulse clusters detected within a preset analysis period .

[0042] In some embodiments, the difference response matrix can be represented as:

[0043] In the formula: This means that the waveform characteristics after the perturbation action are differiated from the waveform characteristics before the perturbation action.

[0044] The second sub-stage is the airflow perturbation stage.

[0045] During the airflow perturbation stage, the gas flow rate or direction is changed by solenoid valves or mass flow controllers, thereby applying airflow perturbation to the plasma processing process.

[0046] Subsequently, voltage, current, and charge waveforms were acquired after the air gap perturbation. Based on the voltage, current, and charge waveforms before and after the air gap perturbation, the differential response matrix before and after the air gap perturbation was obtained. The formation process of the differential response matrix has been explained in the air gap perturbation stage, and the same principle applies here, so it will not be repeated.

[0047] The third sub-stage is the mechanical disturbance stage.

[0048] During the mechanical disturbance stage, the local distribution of contaminated soil can be altered by slight turning, vibration, or compaction, thereby applying mechanical disturbance to the plasma treatment process.

[0049] Subsequently, voltage, current, and charge waveforms were acquired after the mechanical perturbation. Based on the voltage, current, and charge waveforms before and after the air gap perturbation, the differential response matrix before and after the mechanical perturbation was obtained. .

[0050] The fourth sub-stage is the parking stage.

[0051] During the parking phase, the high-voltage output can be paused and then restarted to record the recovery response, thereby applying a parking disturbance to the plasma processing process.

[0052] Subsequently, voltage, current, and charge waveforms were acquired after the parking recovery operation. Based on the voltage, current, and charge waveforms before and after the air gap perturbation operation, the differential response matrix before and after the parking recovery diagnostic operation was obtained. .

[0053] During the aforementioned plasma processing, voltage waveforms were simultaneously acquired. Current waveform and charge waveform The sampling frequency can be adjusted according to the time scale of the discharge pulse, and should meet the requirement of being able to distinguish individual micro-discharge pulses.

[0054] In some embodiments, if there is a propagation delay in the voltage, current, and charge paths, the delay is calibrated through a standard capacitive load or no-load discharge path. Phase compensation is performed on the acquired signals.

[0055] In some embodiments, the acquired current waveform can be processed to remove displacement current and improve the accuracy of the acquired waveform characteristics.

[0056] Current It can be expressed as displacement current. With micro-discharge pulse current The sum of .

[0057] Displacement current can be approximated as ,in The reference capacitor is used. This allows us to obtain the micro-discharge pulse current. By subtracting displacement current, the number, amplitude, and phase of micro-discharge pulses can be extracted more accurately.

[0058] In some embodiments, the waveform can be divided into single-cycle or half-cycle data segments based on the voltage zero point, peak point, or power trigger signal. A reference feature vector is established based on the initial sample processing stage or low-energy discharge stage. The normalized features will be calculated for each subsequent window. , , , and This allows it to adapt to different electrode sizes, different input powers, different air gaps, and different initial soil conditions.

[0059] In some embodiments, in step S103, voltage waveforms, current waveforms, and charge waveforms before and after each active diagnostic action are collected, and corresponding waveform tensors, differential response characteristics, and physical response exponents are constructed based on the paired waveforms before and after the active diagnostic action is executed.

[0060] In some embodiments, the waveform tensor includes voltage, current, and charge waveforms before the active diagnostic action is executed, voltage, current, and charge waveforms after the action is executed, and differential signals between the waveforms before and after the action.

[0061] In some embodiments, the differential response characteristics include at least three of the following: discharge voltage difference, micro-discharge pulse number difference, pulse peak current difference, long-tail current difference, Lissajous figure area difference, equivalent capacitance difference, pulse phase difference, single-cycle energy difference, high-frequency energy ratio difference, and pulse cluster number difference.

[0062] In some embodiments, the physical response index includes at least two of the following: the conductive channel retention index, the shading migration index, the compaction response index, the organic self-evolution trajectory index, and the drift-related index.

[0063] In some embodiments, the operating parameters include at least three of the following: discharge power, frequency, duty cycle, air gap, gas flow rate, sample layer thickness, and sample water content.

[0064] (1) Identification and handling of conductive particles Typical objects: conductive particles include metal debris, high-salt wet clumps, local wet spots with extremely high water content, conductive spots formed by carbonization during discharge, and doped conductive mineral particles.

[0065] Waveform characteristics: starting and discharging voltage Significantly reduced, for example <0.75. Single-pulse peak current Significantly increased, for example >2.0. Number of micro-discharge pulses Reduced, but single pulse energy or pulse charge Increase, for example <0.6, An increase in current with a long tail indicates a prolonged period of partial conduction. Local voltage waveform collapse or clipping can be detected using the voltage collapse index. Characterization. Lissajous figure Sharp corners, local flattening, or abrupt changes in slope may occur.

[0066] In some embodiments, the voltage slump index for:

[0067] In the formula: This represents the theoretical reference voltage or benchmark fitting voltage corresponding to the current phase position or the current sampling time under conditions of no local conduction, arcing, or abnormal load disturbance. This indicates that in the current analysis period, in relation to... The actual discharge voltage measured at the same sampling time or the same phase position.

[0068] like: This indicates that there may be partial conduction or quasi-arc.

[0069] In some embodiments, a conductive channel retention index can be defined based on the conductivity of the conductive channel after an air gap perturbation. :

[0070] In the formula: This represents the long-tailed current integral obtained within the same phase and the same long-tailed analysis interval after the air gap perturbation action is performed. The long-tailed current integral Δd, obtained by integrating the absolute value of the pulse current after deducting the displacement current within a preset long-tailed analysis interval before the air gap perturbation action is executed, represents the air gap perturbation amount, which is equal to the difference between the air gap width after perturbation and the air gap width before perturbation. This represents the initial air gap width before air gap perturbation.

[0071] The conductivity channel retention index is used to characterize the degree to which the long tail current does not decay proportionally after the air gap is increased or the power is reduced during detection.

[0072] Conductive particle criterion: When the discharge voltage decreases, the peak current increases, and the long-tail current strengthens during the natural discharge phase, the system enters the air gap perturbation diagnosis stage or the low-energy detection discharge stage. If, after a slight increase in air gap or a decrease in discharge power, the long-tail current integral, voltage collapse, or local sharp angles in the Lissajous figure do not decrease proportionally with the increase in air gap or decrease in power, and the conductivity channel maintenance index exceeds the set range, then it is determined to be a disturbance caused by conductive particles, local wet clumps, high-salt conductive points, or carbonized conductive points.

[0073] (2) Identification and treatment of insulating hard impurities Typical objects: Insulating hard impurities include glass flakes, pebbles, ceramic flakes, large-particle gravel, and hard mineral blocks with low moisture content.

[0074] Waveform characteristics: starting and discharging voltage Elevation, for example >1.15. Number of micro-discharge pulses Reduce, for example <0.75. The peak value of a single pulse does not change drastically, typically 0.7 < <1.5. The equivalent capacitance changes slowly or abruptly, for example, | -1|>0.10. The area of ​​a Lissajous figure decreases or the slope changes, for example... <0.85. The pulse phase shift indicates that the discharge is delayed.

[0075] In some embodiments, a pulse phase center is defined. for: In the formula: Indicates the first The phase of a micro-discharge pulse.

[0076] when: This indicates a delayed discharge, which may be due to insulation obstruction or a localized increase in the air gap; among which, This indicates the preset threshold.

[0077] In some embodiments, an abnormal migration index after mixing is defined. : In the formula: This indicates abnormal characteristics of obstruction before mixing. This indicates the abnormal characteristics of shading after mixing. These abnormal characteristics can be composed of discharge delay, pulse reduction, and Lissajous area decrease.

[0078] Secondary estimation: Secondary estimation can be performed based on the persistence of the anomaly, the response after stirring, and the trend of equivalent capacitance change. Large pieces of glass / stone flakes usually exhibit intermittent step anomalies, which may suddenly disappear or shift after stirring; stones / ceramic blocks have stronger anomaly persistence, with a significant shift in pulse phase; and large amounts of sand particles change more slowly, closer to the overall baseline.

[0079] Criteria and Handling: When the discharge phase is characterized by an increase in discharge voltage, a decrease in the number of pulses, a shift in pulse phase, and a decrease in the Lissajous figure area, the system performs a slight agitation or air gap fine-tuning diagnosis. If the above-mentioned shielding anomalies migrate, weaken, or disappear after agitation, without significant long-tail current and voltage collapse, it is determined to be a shielding disturbance caused by hard insulating impurities; if the anomaly is not sensitive to agitation but is sensitive to sieving, it is further determined to be a large-diameter hard impurity or a persistent shielding object.

[0080] Secondary estimation of insulating hard impurities: After the primary classification model identifies the current disturbance as an insulating hard impurity, a secondary estimation is performed based on anomalous suddenness, persistence, agitation response, phase hysteresis, equivalent capacitance drift, and pulse phase distribution. This secondary estimation, belonging to engineering classification estimation, is used to determine whether the impurity is closer to a plate-like hard material, a blocky hard material, or a high sand content state, rather than requiring the determination of the impurity's chemical elemental composition as a necessary condition.

[0081] This application sets up a fast time window and a slow time window. The fast time window includes one voltage half-cycle, one voltage cycle, or multiple consecutive discharge cycles, and is used to extract transient features such as discharge voltage, number of micro-discharge pulses, pulse peak value, long-tail current, pulse phase, pulse cluster, Lissajous figure area, and single-cycle discharge energy. Preferably, the fast time window includes 10 to 100 consecutive discharge cycles.

[0082] The slow time window consists of feature vectors output from multiple consecutive fast time windows, used to analyze the evolution trends of equivalent capacitance, Lissajous figure area, micro-discharge pulse number, long-tail current, and high-frequency energy ratio with processing time. The preferred length of the slow time window is 5–30 s, and the preferred update step size is 1–5 s; for organic impurities with longer evolution times, such as leaves, plant fibers, wood chips, or roots, a longer slow time window of 30–120 s can also be set.

[0083] In one implementation, the slow time window length is 10 s, and the window step size is 2 s; the system extracts the fast time feature sequence within the most recent 10 s every 2 s, and calculates the rate of change and trend slope of each feature through least squares fitting. The window length and step size can be adjusted according to the discharge frequency, material handling time constant, and duration of active diagnostic actions.

[0084] Insulation abnormality amplitude: for the first Construct an insulation anomaly feature vector using a slow time window:

[0085]

[0086]

[0087]

[0088]

[0089]

[0090] In the formula: the parameter with subscript 0 is the baseline value of the current batch. ε represents the standard deviation of the parameters corresponding to the baseline stage, and ε is a positive number to avoid a denominator of zero.

[0091] In some embodiments, the first The comprehensive insulation anomaly amplitude of each window Defined as:

[0092] in, The weight matrix for each feature can be determined from known impurity calibration data.

[0093] Step mutation index: When a sheet of glass or stone enters or leaves the discharge coverage area, it usually manifests as an abnormal and sudden appearance or disappearance.

[0094] In some embodiments, a step change index is defined. for:

[0095] in, The higher the value, the closer the anomaly is to local, sudden, and location-sensitive occlusion.

[0096] In some embodiments, an anomaly persistence index is defined. for:

[0097] In the formula: To analyze the total number of windows, As the baseline anomaly threshold, This is an indicator function. The higher the value, the more persistent the anomaly, and the closer it is to a stable blockage caused by pebbles, ceramic blocks, or high sand content.

[0098] In some embodiments, the average insulation anomaly vectors before and after mixing are respectively:

[0099]

[0100] Then the mixing migration index Defined as:

[0101] The recovery index of overmixing Defined as:

[0102]

[0103]

[0104] It should be noted that the mixing migration index , recovery index Related index of mixing K stir Each represents a different technical meaning. Among them, The distance between the multidimensional abnormal feature vectors before and after mixing is used to characterize the degree of migration or reconstruction of abnormal features. The value is determined by the percentage decrease in the average abnormal amplitude before and after mixing, and is used to characterize the degree of improvement of the abnormal state by a single mixing action.K stir The maximum normalized delay correlation coefficient between the abnormal waveform amplitude sequence and the turning action signal within a continuous time window is used to characterize whether a stable temporal correspondence exists between abnormal changes and the turning action. These three parameters are not the same and are input into the hierarchical classification model as different physical features.

[0105] Flakes of glass or stone are more prone to displacement after being stirred, therefore and Typically higher; stones or ceramic blocks that are stably embedded in a sample usually have a weaker stirring response.

[0106] In some embodiments, a phase lag index is defined. for:

[0107] A higher discharge rate and a longer duration indicate a long-term lag in discharge, more closely resembling that of boulders, ceramic blocks, or stable hard mineral blocks.

[0108] In some embodiments, in continuous Within a slow time window, the normalized air gap equivalent capacitance... Perform least squares fitting:

[0109] Subsequently, the slow drift index of the equivalent capacitance is defined. for:

[0110] High sand content typically doesn't manifest as a sudden movement of a single obstruction, but rather as a slow change in the overall equivalent dielectric state. higher and Relatively low.

[0111] In some embodiments, the phase of a micro-discharge pulse within an analysis window is divided into The interval, the first The pulse ratio of each interval is Define the normalized pulse phase entropy:

[0112] A large number of dispersed sand grains will cause multiple local dielectric interfaces and relatively dispersed micro-discharge sites, whose pulse phase distribution is usually wide, therefore It can be used to assist in characterizing dispersed sand particle disturbances.

[0113] In some embodiments, a second-order estimated feature vector is constructed:

[0114] In some embodiments, the score for sheet glass or stone flakes is defined as:

[0115] In some embodiments, the score for the stone or ceramic block class is defined as:

[0116] In some embodiments, the score for the high sand content category is defined as:

[0117] Among them, each These are the model coefficients obtained through calibration experiments using known impurities. The above formulas clearly define the recommended directions of action for each physical quantity; in another implementation, the signs of the coefficients may not be predetermined, and the training process may learn them automatically.

[0118] In some embodiments, the probability of the corresponding secondary category is:

[0119] in, Indicates a sheet of glass or stone; Indicates pebbles or ceramic blocks; Indicates high sand content.

[0120] In some embodiments, the model coefficients are obtained through multi-class logistic regression with regularization terms:

[0121] In the formula: This represents the true label of the second-level class k corresponding to the nth training sample. This represents the regularization coefficient, used to control the balance between model fitting error and parameter complexity. In some embodiments, when the second-level classification has the highest probability of satisfying And the difference between the highest probability and the second highest probability satisfies At that time, the corresponding secondary estimation result is output. In one implementation, it can be set to... , .

[0122] If the above conditions are not met, the output will be "Insulating hard mixed impurities" or "Insulating hard impurity type to be confirmed", and a second mixing diagnosis or sieving verification will be performed, without forcibly giving a definite conclusion about glass, stones or sand.

[0123] To enable the secondary estimate to output not only the category but also the size of the impurity perturbation, a regression estimation header can be added after the secondary classification model. Let the latent vector of the secondary classification feature or integrated model be... Define the equivalent content of impurities: ,in, This represents a normalized estimate of the mass fraction, volume fraction, or projected coverage of impurities.

[0124] In some embodiments, the equivalent particle size of the impurity is defined as:

[0125] For flaky impurities, the equivalent blocking area can also be output:

[0126] The model is trained using calibration experiments with known added mass, known particle size, or known projected area. The output is an estimate of the equivalent perturbation size and is not equivalent to an accurate count of the actual quantity of each impurity in the reactor.

[0127] (3) Identification and treatment of large-sized aggregates Typical objects: Large-sized aggregates include unbroken contaminated soil clumps, oil-based drill cuttings aggregates, clay lumps, surface crusts, and crust layers formed during the drying process.

[0128] Waveform characteristics: Two or more distinct pulse clusters appear within the same half-cycle, i.e. ≥2. An increased pulse time interval indicates a discontinuous partial discharge channel. Discharge voltage variation coefficient. An increase indicates instability in the local air gap. The equivalent capacitance exhibits periodic or slow-time drift, reflecting thermal drying, collapse, or breakage of the aggregates. Increased single-cycle energy fluctuations, for example... Exceeding the preset threshold.

[0129] In some embodiments, a waveform improvement index is defined to describe whether the waveform improves before and after compaction / vibration. :

[0130] in: This refers to the compaction displacement or the intensity of the compaction action.

[0131] Criteria and Handling: When multiple pulse clusters, increased discharge voltage fluctuations, and enhanced single-cycle energy fluctuations occur during the natural discharge phase, the system performs compaction or vibration diagnosis. If, after compaction or vibration, the number of multiple pulse clusters decreases, the single-cycle energy variation coefficient decreases, and the Lissajous figure recovers from a multi-segment to a more stable closed shape, it is determined to be large-sized agglomerates, soil clods, or crust layer disturbance; if the abnormality remains unchanged after compaction but abnormal migration occurs after turning over, it is reclassified as insulating hard impurities.

[0132] The handling strategies include: using low-power pretreatment in the early stage to avoid surface crusting caused by direct high power; if the treatment is insufficient, start stirring, compaction or mechanical crushing to reduce the sample thickness and restore discharge uniformity; after the agglomerates collapse, re-establish the reference waveform and restore normal processing parameters.

[0133] (4) Identification and treatment of the state machine evolution of organic drift matter Organic drift is one of the most important differentiating features of this invention. This invention further categorizes it into leaf / plant fiber, lightweight organic fragments, and wood / root / stem particles.

[0134] For impurities such as leaves / plant fibers, the judgment is not based solely on whether the current peak value at a certain moment exceeds the threshold, but rather on the state transition sequence formed by multiple slow time windows.

[0135] The system calculates the rate of change of equivalent capacitance, the rate of change of Lissajous figure area, the rate of change of micro-discharge pulse number, the rate of change of high-frequency energy, and the rate of change of long-tail current, respectively. When a sequence of "slow decrease in equivalent capacitance - decrease in Lissajous figure area - decrease in pulse number - followed by a short-term increase in long-tail current - then entering a stable decay" appears in multiple consecutive windows, it is determined that the organic impurity has undergone a process of drying shrinkage, carbonization and conductivity, and crushing and ashing, thus identifying it as a leaf or plant fiber impurity.

[0136] Table 1: State Evolution Process of Organic Drift Matter

[0137] See Table 1 for a detailed illustration of the state evolution process of organic drift materials.

[0138] This invention defines an organic self-degradation evolution index. It is used to comprehensively represent the slow-time evolution of equivalent capacitance, Lissajous area, number of micro-discharge pulses, and carbonization burst index.

[0139] when When a sequence of "slow drift - sudden conductivity peak - recovery to stability" is presented, it is identified as a leaf / plant fiber impurity.

[0140] In some embodiments, the organic self-degradation evolution index Represented as:

[0141] in: Indicates the carbonization burst index; , , , This represents the weighting coefficient.

[0142] In some embodiments, continuous carbonization probability Represented as:

[0143] In the formula: Represents the normalized value of the current; This represents the standardized value of the long-tailed current integral; This represents the normalized value of voltage sag; This represents the standardized value of the high-frequency energy ratio; This indicates the standardized value of the discharge voltage; This indicates the conductivity channel retention index.

[0144] The condition of leaves is no longer based on a single factor. The decision is made by adopting a four-state hidden state model:

[0145] The observation vector corresponding to each state is:

[0146] State transition satisfies It allows the same state to be maintained and to be transferred in the direction of "moisture-containing flexibility - drying shrinkage - carbonization conductivity - crushing and ashing".

[0147] In some embodiments, the Viterbi algorithm is used to solve for the most likely state sequence:

[0148] Therefore, the identification results of leaf or plant fiber impurities are derived from the state evolution sequence of multiple windows.

[0149] Lightweight organic debris: Lightweight organic debris does not exhibit continuous unidirectional evolution, but rather shows a delayed correlation between anomalous waveforms and airflow or agitation. The system calculates the delayed correlation coefficients between anomalous waveform sequences and airflow disturbance sequences, as well as agitation sequences, by changing the airflow rate or pausing agitation. When anomalies occur intermittently and have a stable delayed correlation with airflow or agitation, and do not satisfy the continuous state transition sequence of leafy organic matter, they are identified as drifting materials such as paper scraps, plastic films, or lightweight wood chips.

[0150] Lightweight organic debris includes plastic films, paper scraps, foam fragments, and lightweight wood chips. These impurities readily enter and leave the discharge region under the influence of airflow, thermal buoyancy, agitation, or electric fields. Their main characteristic is spatial instability, rather than monotonous degradation. Anomalies occur intermittently and can be characterized by the intermittent duty cycle index, which is the ratio of the anomaly duration to the total observation window time. The intermittent duty cycle index for lightweight drifting materials typically ranges from 0.1 to 0.6.

[0151] To quantitatively characterize the correlation between abnormal discharge states and airflow and tumbling actions, an abnormal waveform amplitude sequence is first constructed. For the first... Using a fast time window, at least three of the following parameters are extracted: discharge voltage, number of micro-discharge pulses, pulse peak current, long-tail current integral, Lissajous figure area, air gap equivalent capacitance, pulse phase center, single-cycle discharge energy, high-frequency energy ratio, and number of micro-discharge pulse clusters, to construct a window feature vector. .

[0152] Standardize relative to the current batch baseline state: .

[0153] In the formula: Indicates the first The first time window Phase discharge characteristics; Indicates the current batch baseline stage number The average value of each feature; Indicates the current batch baseline stage number The standard deviation of the feature; This indicates the avoidance of small positive numbers with a denominator of zero.

[0154] No. The comprehensive abnormal waveform amplitude for each time window is defined as follows: .

[0155] In the formula: Indicates the first Standardized anomaly feature vectors for each time window; The weight matrix representing the abnormal features can be an identity matrix or determined based on known impurity calibration data. Indicates the first The comprehensive abnormal waveform amplitude of each time window.

[0156] continuous The windows constitute a sequence of abnormal waveform amplitudes: .

[0157] The airflow disturbance signal is defined as: .

[0158] In the formula: Indicates the first Normalized airflow disturbance signal for each time window; Indicates the first The actual gas flow rate within a time window; This represents the reference gas flow rate before the airflow perturbation.

[0159] When gas flow rate is not continuously measured but only the solenoid valve or airflow switching status is recorded, it can also be set as follows:

[0160] For the number of delay windows Define the airflow delay correlation coefficient:

[0161] In the formula: Indicates delay The correlation coefficient between the abnormal waveform amplitude and the airflow disturbance signal within a window; This represents the average amplitude of the abnormal waveform within the corresponding calculation interval; This represents the average value of the airflow disturbance signal within the corresponding calculation interval; This represents the total number of windows contained within the slow-time analysis interval.

[0162] The airflow-related index is defined as:

[0163] In the formula: This indicates the preset search range for airflow response delay; The value range is from 0 to 1, where, The larger the value, the stronger the time correlation between the abnormal waveform and the airflow movement.

[0164] The corresponding airflow response delay time is:

[0165] in: , This indicates the update time step between adjacent feature windows.

[0166] The stirring action signal is defined as:

[0167] When the stirring speed, vibration amplitude, or duration of the action can be obtained... It can also be the normalized continuous motion intensity.

[0168] For the number of delay windows Define the correlation coefficient of mixing delay:

[0169] In the formula: Indicates delay The correlation coefficient between the abnormal waveform amplitude and the stirring action signal during each window; This represents the average value of the stirring action signal within the corresponding calculation interval.

[0170] The relevant index for mixing is defined as follows:

[0171] In the formula: This indicates the search range for the delay in response to the stirring action; The value range is from 0 to 1; where, The larger the value, the stronger the time correlation between the abnormal waveform change and the actions of stirring, vibration, or material redistribution.

[0172] The corresponding stirring response delay time is:

[0173] in:

[0174] Anomalies are related to airflow pulsations or tumbling motions, and can be addressed using the drift correlation index. Characterization.

[0175] If the occurrence of an anomaly is related to the airflow pulsation cycle or tumbling cycle, the drift probability increases.

[0176] Define the drift correlation index: or ; in: This indicates a gas flow rate or gas flow pulsation signal; This indicates a stirring or mixing action signal; This represents the sequence of abnormal waveform amplitudes.

[0177] It does not exhibit the continuous, monotonous evolution of leaves—"slow drift-carbonization-stabilization." Unlike leaves, lightweight drifting objects do not show a continuous... , , Instead of a monotonically decreasing or "carbonization-recovery" sequence, it exhibits sudden appearance, sudden disappearance, and reappearance. High-frequency jitter is enhanced, but the peak value of a single pulse does not necessarily increase significantly. When the abnormal intermittent duty cycle index... Within the preset interval range, airflow-related index Related index of mixing At least one of them is higher than the corresponding calibration threshold, and the organic self-evolution trajectory index When a continuous drying-carbonization-ashing state transition sequence is not observed, the classification probability of airflow-related lightweight drift particles is increased. Treatment strategies include reducing gas flow rate, changing the inlet direction, pausing agitation, reducing peak power, and increasing the duty cycle of low-power continuous discharge to cause slow contraction, fixation, or degradation; if the anomaly persists, the screening process should be stopped.

[0178] Wood chips / roots: Wood chips and roots are thicker than leaves, have low driftability, but dry and carbonize slowly. They are characterized by a longer evolutionary time constant than leaves. The changes are slow, with infrequent but prolonged carbonization bursts, and a potential long current tail may appear in the later stages. The treatment strategy involves phased processing, low-power preheating, and continued processing after stirring. If stable conductive hot spots form in the later stages, the affected parts should be screened or crushed.

[0179] See Figure 3 This specifically illustrates the waveform change trends of different types of impurities after a disturbance is applied.

[0180] In some embodiments, in step S104, the waveform tensor, differential response features, physical response index, and plasma processing condition parameters corresponding to each active diagnostic action are input into a pre-trained multi-perturbation branch hierarchical classification model; the multi-perturbation branch hierarchical classification model extracts the deep waveform features corresponding to each active diagnostic action, determines the fusion weight based on the discharge response corresponding to each active diagnostic action, and fuses the fused waveform features with the physical response features to output the primary category, secondary category, risk level, and classification confidence of the impurity.

[0181] This invention, based on the physical interpretation using the conductive channel retention index, shading migration index, compaction response index, organic self-evolution trajectory index, and drift correlation index, further constructs a multi-perturbation branch hierarchical classification model guided by physical indicators. This multi-perturbation branch hierarchical classification model does not directly determine the impurity type using a single threshold, but simultaneously processes the original discharge waveform, differential response characteristics, physical response index, and discharge condition parameters before and after active diagnostic actions, outputting the impurity's primary category, secondary category, risk level, and classification confidence level.

[0182] In some embodiments, the input features of the multi-perturbation branch hierarchical classification model used in this application for:

[0183] In the formula: Indicates the first The waveform tensor corresponding to the diagnostic action. Indicates the first The differential response vector corresponding to the diagnostic action. Represents the physical response exponent vector. This represents the discharge condition parameter vector. This is a mask for proactive diagnostic actions, used to indicate whether the current sample has performed the corresponding diagnostic action. Unexecuted diagnostic branches are removed from subsequent fusion calculations using this mask.

[0184] In some embodiments, for any active diagnostic action ,definition: ,in Indicates air gap perturbation; Indicates a slight disturbance in the airflow; Indicates mechanical perturbation; Indicates that the parking area has been restored; This indicates a low-energy detection discharge.

[0185] In some embodiments, waveform tensor The extraction process is as follows: First, voltage, current, and charge waveforms were collected before and after the active diagnostic action:

[0186]

[0187] And construct the corresponding differential waveform: .

[0188] Subsequently, the waveforms of each period were resampled to the same... 10 sampling points, and select continuous If the discharge cycle is used as a model input sample, then the first discharge cycle... Waveform tensor corresponding to diagnostic actions for:

[0189] in, It includes voltage, current, and charge before the disturbance, voltage, current, and charge after the disturbance, as well as the differential signal of the three, for a total of 9 input channels.

[0190] In some embodiments, in addition to the original waveform, a corresponding differential response vector is also input. :

[0191] In the formula: Indicates the first The differential response vector corresponding to each active diagnostic action (also known as a perturbation action).

[0192] In some embodiments, a physical response exponent vector may be further input. Physical response exponential vector Defined as:

[0193] In the formula, It represents the conductivity channel retention index, used to characterize the degree of retention of long-tail current and local conductivity response after air gap perturbation or low-energy probe discharge; The occlusion migration index is used to characterize the degree of migration or reconstruction of multidimensional abnormal feature vectors before and after mixing. It represents the compaction response index, used to characterize the degree of improvement in pulse clusters, discharge voltage fluctuations, and single-cycle energy fluctuations before and after compaction or vibration; It represents the organic self-evolution trajectory index, used to characterize the drying, shrinkage, carbonization, and ashing evolution processes of organic impurities within multiple slow time windows; The airflow correlation index is obtained by calculating the maximum normalized delay correlation coefficient between the abnormal waveform amplitude sequence and the gas flow rate change signal, airflow direction switching signal or pulse intake signal within a preset delay range. The tumbling correlation index is obtained by calculating the maximum normalized delay correlation coefficient between the abnormal waveform amplitude sequence and the tumbling, vibration or other mechanical disturbance action signal within a preset delay range.

[0194] In some embodiments, a discharge condition parameter vector may be further input. Discharge condition parameter vector Defined as:

[0195] In the formula: Indicates discharge power; Indicates the discharge frequency; Indicates the duty cycle; Indicates the discharge air gap; Indicates gas flow rate; Indicates the thickness of the sample layer; This indicates the moisture content of the sample or its equivalent moisture content.

[0196] In some embodiments, the multi-perturbation branch hierarchical classification model includes: The waveform deep feature extraction branch is used to extract waveform deep features based on the waveform tensor of the active diagnostic action; The physical feature branch is used to extract physical feature representations based on differential response characteristics, physical response index, and operating parameters. The attention fusion module is used to obtain attention intermediate features from the waveform deep features, differential response features, and operating condition features of each active diagnostic action; it receives the active diagnostic action mask and determines the fusion weight of each type of active diagnostic action based on the attention intermediate features and the active diagnostic action mask; then, it uses the fusion weight to perform weighted fusion of the attention intermediate features of several active diagnostic actions to obtain the active disturbance waveform features. The first-level classification head is used to determine the physical prior classification score based on the physical response index characteristics; determine the data-driven classification score based on the active perturbation waveform characteristics and physical feature representation; calculate the gating coefficient based on the waveform signal quality, the completeness of the active diagnostic action and the stability of the current batch baseline waveform, and fuse the physical prior classification score and the data-driven classification score based on the gating coefficient to determine the first-level category and first-level classification confidence of the impurity. The secondary classification head is used to determine the secondary category, confidence level, and risk level of impurities based on the waveform and physical characteristics of the active disturbance.

[0197] In some embodiments, the deep waveform feature extraction branch is used for the waveform tensor formed by each active diagnostic action. A one-dimensional convolutional network with shared parameters and a bidirectional gated recurrent network are used to extract deep waveform features. Specifically, it is expressed as:

[0198] In the formula: This indicates the deep feature extraction branch of the waveform. This represents a bidirectional gated loop network. This represents a one-dimensional convolutional network.

[0199] In some embodiments, the one-dimensional convolutional network includes three convolutional layers with kernel sizes of 7, 5, and 3, and output channels of 32, 64, and 128, respectively. The convolution result is input into a bidirectional gated recurrent unit to simultaneously extract the local morphology of a single-cycle micro-discharge pulse and the evolutionary relationship between consecutive cycles.

[0200] The waveform information learned by the deep feature extraction branch of this waveform includes at least: the shape and peak leading edge features of the micro-discharge pulse; the internal structure of the pulse cluster and the changes in pulse interval; the attenuation pattern and long tail persistence after the current peak; the local drift of the waveform phase and the period migration; the dynamic changes in the formation process of the Lissajous figure; and the nonlinear differences in the waveform before and after the active diagnostic action.

[0201] The above network structure is one implementation method provided by this application. This application may also use temporal convolutional networks, long short-term memory networks, Transformer temporal encoders or other feature extraction networks that can process discharge waveform sequences as deep feature extraction branches of waveforms.

[0202] In some embodiments, the physical feature branch is used to input the differential response vector, physical response index, and operating condition parameters into a fully connected feature extraction network to obtain a physical feature representation. : First, the differential response vector, physical response index, and operating parameters are concatenated to form the physical feature input vector. , is represented as:

[0203] Input the physical features into the vector Inputting the data into a multilayer perceptron yields physical feature representations. :

[0204] In the formula: This represents a multilayer perceptron.

[0205] In some embodiments, the multilayer perceptron includes two fully connected layers, and its computation process is as follows:

[0206] In the formula: , For model weights; , For bias; This is a non-linear activation function. This branch is used to preserve the interpretable correlation between active diagnostic actions and the physical response of the discharge.

[0207] In some embodiments, since different impurities have different sensitivities to different active diagnostic actions, the present invention automatically determines the contribution of each diagnostic action to the identification result by attention weight.

[0208] The attention fusion process of the attention fusion module is as follows: No. Attention intermediate of each active diagnostic branch for:

[0209] In the formula: Indicates the first Deep waveform features corresponding to each active diagnostic action (also known as a perturbation action). Indicates the first Differential response features corresponding to each active diagnostic action (also known as a perturbation action). Indicates the characteristics of the operating conditions. Indicates the first The bias corresponding to each active diagnostic action (also known as a perturbation action). express The weight, express The weight, express The weight.

[0210] In some embodiments, the first Attention weights corresponding to each proactive diagnostic action for:

[0211] In the formula: This represents the valid mask for the type a active diagnostic action; when the active diagnostic action has been executed and the corresponding waveform data meets the quality requirements. Select 1 if the value is 1, otherwise select 0. This represents the attention parameter vector obtained through model training. This represents the transpose of the attention parameter vector. This represents the inner product of the attention parameter vector and the attention intermediate feature vector corresponding to the a-th type of active diagnostic action. Indicates the first term in the summation term Effective mask for proactive diagnostic actions.

[0212] Deep waveform features of several active diagnostic actions After fusion, the active perturbation waveform characteristics are obtained. for:

[0213] Finally, the comprehensive feature vector is:

[0214] Therefore, the model can automatically increase the weight of effective diagnostic actions based on the actual waveform, rather than pre-determining that a certain diagnostic action necessarily corresponds to a certain impurity.

[0215] In some embodiments, typical weight variations for different categories can be interpreted as follows: in the identification of conductive particles, the air gap perturbation and low-energy detection branches typically contribute more; in the identification of insulating hard impurities, the mechanical stirring branch typically contributes more; in the identification of large-sized agglomerates, the compaction or vibration branch typically contributes more; in the identification of lightweight organic drifting matter, the airflow perturbation branch typically contributes more; and in the identification of leafy organic impurities, the low-energy detection and parking recovery branches typically contribute more.

[0216] The phrase "usually has a high contribution" is only used to explain the physical meaning of the model. The actual attention weights are learned from the training samples and are not used as a fixed threshold constraint.

[0217] In some embodiments, the classification process of the primary classification head is as follows: According to the physical response index Calculate the physical prior classification score:

[0218] Data-driven classification scores are calculated based on deep waveform features:

[0219] The fusion gating coefficients are generated based on the waveform signal-to-noise ratio, the completeness of the diagnostic action, and the stability of the reference waveform.

[0220] in, As a signal quality indicator, This is the Sigmoid function.

[0221] Finally, the probability of the first-level classification is:

[0222] The primary categories include:

[0223] In this fusion approach, the physical discriminant index serves as a soft prior for the model, rather than a hard constraint that must be satisfied simultaneously. Even if a certain physical index is affected by noise, soil batches, or changes in operating conditions, the model can still complete a comprehensive identification using the original waveform, other diagnostic responses, and operating parameters.

[0224] In some embodiments, the multi-perturbation branch hierarchical classification model further includes a secondary classification head for classifying active perturbation waveform features. and physical characteristics representation Calculate the probability of secondary classification of impurities.

[0225] In some embodiments, the classification process of the secondary classification head is as follows: After the primary category is completed, the corresponding secondary category header is called based on the primary category:

[0226] in, This indicates the primary category.

[0227] Table 2: Types of Secondary Categories

[0228] See Table 2 for a detailed breakdown of the types of secondary categories.

[0229] In some embodiments, the secondary categories include at least the contents shown in the table below.

[0230] In some embodiments, the multi-perturbation branch hierarchical classification model can also output risk level probabilities:

[0231] Risk levels include general disturbance, moderate risk, and risk requiring interlocking measures.

[0232] In some embodiments, model training samples can be obtained through experiments involving the mixing of known impurities.

[0233] Each training sample shall record at least the following information: (1) the basic type of soil or oil-based drill cuttings; (2) the primary and secondary categories of impurities; (3) the particle size, mass fraction, volume fraction or projected coverage of impurities; (4) the location and burial depth of impurities in the sample layer; (5) the sample moisture content, oil content, thickness and compaction state; (6) the discharge power, frequency, duty cycle, air gap and gas flow rate; (7) the type, amplitude and duration of active diagnostic action; (8) the voltage, current and charge waveforms before and after the active diagnostic action; and (9) the treatment action and the recovery results after treatment.

[0234] Training samples should cover different impurity sizes, quantities, locations, moisture content, and basic soil types. Training, validation, and test sets should be divided according to material batches. Adjacent waveform windows formed by the same batch should not be simultaneously assigned to the training and test sets to avoid the model only memorizing waveforms from the same batch.

[0235] In some embodiments, the loss function of the multi-perturbation branch hierarchical classification model is: The first-level classification loss is:

[0236] In the formula: This represents the true label of the first-level category corresponding to the nth training sample. It is set to 1 when the nth training sample belongs to the first-level category c, and 0 otherwise. This represents the probability that the model predicts the nth training sample belongs to the first-level category c. Indicates the first-level category Corresponding category weights; Indicates the training sample index. This indicates a first-level category index.

[0237] The second-order classification loss is:

[0238] In the formula: Indicates the first-level category Conditions, No. Each sample corresponds to a secondary category. The true label. The model predicts the first... Each sample belongs to the first-level category. Sub-secondary category The probability of. ω_c,k This represents the category weight corresponding to the second-level category k under the first-level category c.

[0239] The physical prior consistency loss is:

[0240] in, This represents the Kullback-Leibler divergence, used to measure the difference between the physical prior probability distribution and the model's first-level classification probability distribution. This represents the effective physical prior weight corresponding to the nth sample. This represents the first-order category probability distribution vector output by the data-driven hierarchical classification model for the nth sample. This represents the first-order physical prior probability distribution of the nth sample, calculated based on physical response indices such as the conductive channel retention index, shading migration index, compaction response index, organic self-evolution trajectory index, and drift correlation index.

[0241] If both impurity content and equivalent particle size estimation are performed simultaneously, then regression loss is used:

[0242] In the formula: Expressing the request loss. This indicates the true equivalent content of impurities in the training samples, obtained through weighing, volume measurement, or image calibration. This represents the equivalent impurity content predicted by the model. This indicates the true equivalent particle size of impurities obtained through sieving, size measurement, or image calculation. This represents the model-predicted equivalent particle size of the impurities.

[0243] The total loss of the model is:

[0244] in, to This is the loss weighting coefficient.

[0245] In some embodiments, the model reasoning and uncertainty category handling process is as follows: Let the highest probability of primary classification be: .

[0246] The difference between the highest probability and the second highest probability is: .

[0247] when and When the time is right, output the corresponding first-level category and second-level category.

[0248] In one implementation, it can be set , .

[0249] When the above conditions are not met, the model does not force the output of a single impurity name, but instead outputs "mixed disturbance" or "uncertain category," and selects the next active diagnostic action with greater information gain based on the current probability distribution. For example, when the probability of insulating hard impurities is close to that of large-sized agglomerates, compaction diagnosis is performed first; when the probability of leaf-like impurities is close to that of lightweight organic debris, airflow disturbance or parking recovery diagnosis is performed first.

[0250] In some embodiments, in step S105, when the classification confidence level does not reach a preset condition, the next active diagnostic action is selected based on the probability distribution of the current categories, and the process returns to step S102 or S103; when the classification confidence level reaches a preset condition, a treatment strategy is generated based on the impurity primary category, secondary category, and risk level; after executing the treatment strategy, at least one active diagnostic action is executed again to obtain the post-treatment disturbance response; when the post-treatment abnormal response does not decrease to a preset range, the treatment level is upgraded, another treatment action is selected, or a safety interlock is triggered.

[0251] In some embodiments, the treatment strategy employed includes at least one of the following: performing discharge power regulation, discharge frequency regulation, duty cycle regulation, air gap regulation, airflow regulation, turning, compaction, vibration, crushing, screening, magnetic separation, and safety interlock.

[0252] See Figure 4 The paper specifically illustrates the treatment process of a plasma treatment method for contaminated soil based on active perturbation response.

[0253] See Figure 5 It specifically illustrates the treatment strategies provided for different types of impurities.

[0254] In some embodiments, for conductive particles, the disposal strategy may be to reduce the power of the plasma treatment process, increase the air gap size, and re-plasma treat the contaminated soil.

[0255] Specifically, the treatment strategy for conductive particles can be: Slight conductivity disturbance in L1: Discharge power decreases by 10%–30%, duty cycle decreases by 10%–30%, start one stirring and observe 3–10 sampling windows.

[0256] L2 Quasi-arc risk: Immediately reduce voltage or power by 30%–60%, reduce duty cycle to 5%–20%, increase air gap by 0.5–2.0 mm, and start stirring or slight vibration; if equipped with a magnetic separation mechanism, start magnetic separation to remove.

[0257] L3 Continuous Arc Risk: Immediately disconnect the high-voltage power supply, activate the interlock protection, output a "suspected metal / highly conductive wet patch" alarm, perform screening or manual inspection, and re-establish the reference waveform after elimination.

[0258] In some embodiments, for insulating hard impurities, the treatment strategy may be to agitate the contaminated soil, sieve the contaminated soil, adjust the frequency and air gap used in the plasma treatment process, and then treat the contaminated soil with plasma again.

[0259] In some embodiments, for large-sized aggregates, the treatment strategy may be to turn over the contaminated soil, compact the contaminated soil, break up the large-sized aggregates in the contaminated soil, and then treat the contaminated soil with plasma again.

[0260] In some embodiments, for large-sized agglomerates, low-power pretreatment is used in the early stage to avoid surface crusting caused by direct high power; if the treatment is insufficient, stirring, compaction or mechanical crushing is started to reduce the sample thickness and restore the discharge uniformity; after the agglomerates collapse, the reference waveform is re-established and the normal processing parameters are restored.

[0261] In some embodiments, for organic drift, the treatment strategy may be to screen the contaminated soil, adjust the airflow rate and / or direction during the plasma treatment process, and then treat the contaminated soil again with plasma.

[0262] In each plasma treatment process, the impurity identification model provided in this application is used to identify impurities that may exist in the contaminated soil and formulate targeted treatment strategies to improve the treatment effect of contaminated soil.

[0263] In conjunction with the foregoing methods, this application provides some specific embodiments: Example: Identification and Interlocking Protection of Conductive Disturbances from Metal Shavings or High-Salt Wet Clumps Oil-contaminated soil was laid in the DBD reactor with an initial air gap of 10 mm, an input power of 220 W, a frequency of 9 kHz, and an air flow rate of 3 L / min. U(t), I(t), and Q(t) were simultaneously acquired using an oscilloscope. During normal discharge, the reference discharge voltage V was recorded. on0 The number of micro-discharge pulses per cycle, N p0 and single-pulse peak current I peak0 During processing, an "R" appeared in one of the windows. Von =0.62, R I =3.1、R N =0.42, and the current long tail time exceeds the threshold, causing a local collapse in the voltage waveform. The system determines this to be a conductive particle disturbance, level L2. The control system immediately executes a 50% power reduction, a 10% duty cycle reduction, a 1mm increase in the air gap, and one initiation of agitation; if the abnormality persists for three consecutive windows, the high-voltage power supply is cut off and a screening prompt is issued. After agitation, R Von R I R NOnce back to normal, the system gradually restores its original parameters.

[0264] The system briefly increases the air gap from 10 mm to 10.3 mm and reduces the power to 50% of the normal value for low-energy detection. If the long-tail current integral decreases by less than 20%, while the long-tail current integral of a normal, impurity-free sample decreases by more than 60% under the same diagnostic action, then conductivity channel retention is determined to exist.

[0265] Figure 6 This is a structural block diagram of a plasma treatment device for contaminated soil based on active perturbation response, provided as an embodiment of this application. See also... Figure 6 ,include: The dielectric barrier discharge reactor 11 is used to hold the contaminated soil to be treated. An adjustable high-voltage power supply 12 is connected to the dielectric barrier discharge reactor and is used to adjust the discharge power, frequency, duty cycle or peak voltage. Waveform acquisition module 13 is used to synchronously acquire discharge voltage, current and charge waveforms; The active diagnostic execution module 14 includes at least two of the following: an air gap perturbation mechanism, an airflow control mechanism, a mechanical perturbation mechanism, and a parking recovery control unit; The processing execution module 15 includes at least one of the following: an air gap adjustment mechanism, an airflow adjustment mechanism, a turning and mixing mechanism, a compaction or vibration mechanism, a crushing or screening mechanism, a magnetic separation mechanism, and a safety interlock mechanism; The controller 16 is connected to the adjustable high-voltage power supply, the waveform acquisition module, the active diagnostic execution module, and the treatment execution module, respectively. The controller is used to establish the baseline discharge characteristics of the current batch, control the active diagnosis execution module to perform at least two active diagnosis actions, construct differential response characteristics based on the waveforms before and after the active diagnosis actions, input the differential response characteristics into a pre-trained impurity identification model, and control the disposal execution module based on the impurity category, risk level and classification confidence level output by the model.

[0266] Figure 7 This is a structural block diagram of a contaminated soil treatment device provided in one embodiment of this application. See also... Figure 7 Contaminated soil treatment equipment may include Figure 6 The aforementioned plasma treatment device for contaminated soil based on active perturbation response typically includes a processor 21 and a memory 22.

[0267] Processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. Processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 21 may also include a main processor and a coprocessor. The main processor is used to process data in the wake-up state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor used to process data in the standby state. Memory 22 may include one or more computer-readable storage media, which may be non-transitory. Memory 22 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in memory 22 is used to store at least one instruction, which is executed by processor 21 to implement the active perturbation response-based contaminated soil plasma treatment method provided by the method embodiments of this application.

[0268] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A plasma treatment method for contaminated soil based on active perturbation response, characterized in that, include: A low-energy detection discharge was performed on the current batch of contaminated soil to be treated, and reference voltage waveform, reference current waveform and reference charge waveform were collected to establish the reference discharge characteristics of the current batch. During the plasma treatment of contaminated soil, the discharge state is monitored. If the discharge state has an abnormal discharge response relative to the current batch reference discharge characteristics, at least two different active diagnostic actions are executed according to a preset active diagnostic sequence. The active diagnostic actions include air gap perturbation action, airflow perturbation action, mechanical perturbation action, low-energy detection discharge action, and parking recovery action. Data acquisition steps: Collect voltage waveforms, current waveforms, and charge waveforms before and after each active diagnostic action, and then construct waveform tensors, differential response characteristics, and physical response exponents; The waveform tensor, differential response features, physical response index, and operating parameters corresponding to each active diagnostic action are input into a pre-trained multi-perturbation branch hierarchical classification model to obtain the first-level category, second-level category, risk level, and classification confidence of the impurities. If the classification confidence level does not meet the preset conditions, a new active diagnostic action is selected based on the probability distribution of the current categories, and the data acquisition step is returned. If the classification confidence level meets the preset conditions, a treatment strategy is generated based on the primary category, secondary category, and risk level of the impurities. After the treatment strategy is executed, when plasma treatment is performed on the contaminated soil, at least one active diagnostic action is executed again, and voltage waveforms, current waveforms, and charge waveforms before and after the execution of the treatment strategy are collected to determine the degree of abnormal response after the execution of the treatment strategy. If the degree of abnormal response is higher than the preset range, the treatment level is upgraded, and another treatment strategy is selected or a safety interlock is triggered.

2. The method for plasma treatment of contaminated soil based on active perturbation response according to claim 1, characterized in that, The air gap perturbation action includes briefly increasing or decreasing the discharge air gap by a preset distance based on the reference air gap. The airflow perturbation action includes changing the flow rate, flow direction, or pulsation mode of the processed gas; The mechanical perturbation actions include turning, vibrating, compacting, or partially breaking up the contaminated soil. The low-energy detection discharge action includes reducing the discharge power, peak voltage, duty cycle, or pulse duration to a preset ratio of the parameters corresponding to the normal processing stage, while maintaining the ability to generate repeatable micro-discharge pulses. The parking recovery action includes pausing the high-voltage output for a preset time and then restarting the discharge, and collecting the recovery response during the restarting process.

3. The plasma treatment method for contaminated soil based on active perturbation response according to claim 1, characterized in that, The waveform tensor includes voltage, current, and charge waveforms before the active diagnostic action is executed, voltage, current, and charge waveforms after the action is executed, and the difference signal between the waveforms before and after the action. The differential response characteristics include at least three of the following: discharge voltage difference, micro-discharge pulse number difference, pulse peak current difference, long-tail current difference, Lissajous figure area difference, equivalent capacitance difference, pulse phase difference, single-cycle energy difference, high-frequency energy ratio difference, and pulse cluster number difference. The physical response index includes at least two of the following: conductive channel retention index, shading migration index, compaction response index, organic self-evolution trajectory index, and drift correlation index; the drift correlation index includes at least one of the following: airflow correlation index and tumbling correlation index. The airflow correlation index is used to characterize the degree of delay correlation between the abnormal waveform response and the airflow perturbation action, and the tumbling correlation index is used to characterize the degree of delay correlation between the abnormal waveform response and the tumbling, vibration, or other mechanical perturbation action. Operating parameters include at least three of the following: discharge power, frequency, duty cycle, air gap, gas flow rate, sample layer thickness, and sample water content.

4. The plasma treatment method for contaminated soil based on active perturbation response according to claim 3, characterized in that, The multi-perturbation branch hierarchical classification model includes: The waveform deep feature extraction branch is used to extract waveform deep features based on the waveform tensor of the active diagnostic action; The physical feature branch is used to extract physical feature representations based on differential response characteristics, physical response index, and operating parameters. The attention fusion module is used to obtain attention intermediate features from the waveform deep features, differential response features, and operating condition features of each active diagnostic action; it receives the active diagnostic action mask and determines the fusion weight of each type of active diagnostic action based on the attention intermediate features and the active diagnostic action mask; then, it uses the fusion weight to perform weighted fusion of the attention intermediate features of several active diagnostic actions to obtain the active disturbance waveform features. The first-level classification head is used to determine the physical prior classification score based on the physical response index characteristics; determine the data-driven classification score based on the active perturbation waveform characteristics and physical feature representation; calculate the gating coefficient based on the waveform signal quality, the completeness of the active diagnostic action and the stability of the current batch baseline waveform, and fuse the physical prior classification score and the data-driven classification score based on the gating coefficient to determine the first-level category and first-level classification confidence of the impurity. The secondary classification head is used to determine the secondary category, confidence level, and risk level of impurities based on the waveform and physical characteristics of the active disturbance.

5. The plasma treatment method for contaminated soil based on active perturbation response according to claim 4, characterized in that, The training process of the multi-perturbation branch hierarchical classification model includes: Perform at least two active diagnostic actions on contaminated soil samples containing known primary and secondary impurities, and acquire voltage, current, and charge waveforms before and after each active diagnostic action, and form waveform tensors, differential response characteristics, physical response exponents, and operating parameters as sample data; The sample data were divided into training, validation, and test sets according to the batches of contaminated soil. A network model was established, including a waveform deep feature extraction branch, a physical feature branch, a multi-perturbation attention fusion module, a first-level classification head, and a second-level classification head. The network model is trained using first-level classification loss, second-level classification loss, and physical prior consistency loss to obtain a multi-perturbation branch hierarchical classification model.

6. The plasma treatment method for contaminated soil based on active perturbation response according to claim 4, characterized in that, The primary categories include conductive channel type impurities, insulating hard shield type impurities, large-size aggregate type impurities, and organic disturbance type impurities. When the primary category is conductive channel type impurities, the corresponding secondary categories include stable conductive particles, wet salt conductive clusters, and carbonized conductive points; When the primary category is insulating hard shielding impurities, the corresponding secondary categories include sheet-like hard insulators, block-like hard insulators, and dispersed particle enrichment areas; When the primary category is large-sized aggregated impurities, the corresponding secondary categories include ordinary soil lumps, bonded aggregates, and surface crusts. When the primary category is organic disturbance type impurities, the corresponding secondary categories include slow-time organic evolutions, airflow-related light drifts, and wood chips or root and stem organic matter; The model also outputs one of the following risk levels: general disturbance, moderate risk, and risk requiring interlocking measures.

7. The plasma treatment method for contaminated soil based on active perturbation response according to claim 6, characterized in that, When the highest probability of the primary classification is lower than the first confidence threshold, or the difference between the highest probability and the second highest probability of the primary classification is lower than the second confidence threshold, the mixed perturbation or uncertain category is output, and the next active diagnostic action is selected based on the difference between the candidate categories. When the classification probabilities of insulating hard-shielding impurities and large-size agglomerated impurities are similar, compaction or vibration diagnosis should be selected. When the classification probabilities of slow-time organic evolutions and airflow-related lightweight drifts are close, airflow perturbation or parking recovery diagnosis is selected.

8. A plasma treatment device for contaminated soil based on active perturbation response, characterized in that, include: A dielectric barrier discharge reactor is used to hold contaminated soil to be treated. An adjustable high-voltage power supply is connected to the dielectric barrier discharge reactor to adjust the discharge power, frequency, duty cycle, or peak voltage. The waveform acquisition module is used to simultaneously acquire discharge voltage, current, and charge waveforms. The active diagnostic execution module includes at least two of the following: an air gap perturbation mechanism, an airflow control mechanism, a mechanical perturbation mechanism, and a parking recovery control unit; The disposal execution module includes at least one of the following: an air gap adjustment mechanism, an airflow adjustment mechanism, a turning and mixing mechanism, a compaction or vibration mechanism, a crushing or screening mechanism, a magnetic separation mechanism, and a safety interlock mechanism; The controller is connected to the adjustable high-voltage power supply, the waveform acquisition module, the active diagnostic execution module, and the treatment execution module, respectively. The controller is used to establish the baseline discharge characteristics of the current batch, control the active diagnosis execution module to perform at least two active diagnosis actions, construct differential response characteristics based on the waveforms before and after the active diagnosis actions, input the differential response characteristics into a pre-trained impurity identification model, and control the disposal execution module based on the impurity category, risk level and classification confidence level output by the model.

9. A contaminated soil treatment device, characterized in that, Including the plasma treatment device for contaminated soil based on active perturbation response as described in claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one piece of program code, which is executed by a processor to implement the plasma treatment method for contaminated soil based on active perturbation response as described in any one of claims 1 to 7.

Images